Shrey160/LFM3-VL-2B-SFTv2-GGUF
Image-Text-to-Text • 2B • Updated • 103
id stringlengths 10 20 | stage int32 1 3 | embodiment stringclasses 5
values | category stringclasses 13
values | think_mode bool 2
classes | difficulty_tier int32 1 3 | prompt_messages_json stringlengths 110 695 | tools_json stringclasses 1
value | images images listlengths 0 2 | oracle_completion stringlengths 62 1k | world_state_json stringlengths 503 991 | point_predictions_json stringclasses 23
values | expected_answer stringclasses 4
values | oracle_r_total float32 2.7 3.1 | n_tokens int32 65 364 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
s2_wh_nav_009_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>station 1</parameter></function></tool_call><tool_... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 251 | |||
s1_point_001 | 1 | tabletop | pointing | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the red cube. Answer in JSON as a list of points: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}], with x and y as integers normalized to 0-1000."}]}] | <think></think>[{"point_2d": [444, 535], "label": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [{"pred_norm": [444, 535], "gt_px": [284.40914772160704, 256.7825337970678], "mask_hit": true, "zone_ok": true}] | 2.85 | 83 | |||
s2_wh_pt_014 | 2 | wheeled | pointing_nav_pad | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue docking pad at station 2. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [590, 301], "label": "blue docking pad"}] | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station":... | [{"pred_norm": [590, 301], "gt_px": [377.3050463439158, 144.49158942680697], "mask_hit": true, "zone_ok": true}] | 2.85 | 73 | |||
s2_shelf_unblock_009 | 2 | shelf | shelf_unblock | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Slot top_center is currently occupied by red cube. First pick red cube and place_on table to clear top_center, then pick green cube, place_in_slot(top_center), and done.</think><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=place_on><parameter=target>t... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"red cube": "top_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", ... | [] | 3.1 | 218 | |||
s1_unstack_008 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it,... | <think>The red cube sits on top of the green cube. First pick red cube and place on table to uncover green cube, then pick green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "table"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"red cube": "green cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "s... | [] | 2.9 | 176 | |||
s2_wh_pt_000 | 2 | wheeled | pointing_nav_pad | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue docking pad at station 2. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [590, 301], "label": "blue docking pad"}] | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station":... | [{"pred_norm": [590, 301], "gt_px": [377.3050463439158, 144.49158942680697], "mask_hit": true, "zone_ok": true}] | 2.85 | 73 | |||
s2_wh_nav_001_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>station 1</parameter></function></tool_call><tool_... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 251 | |||
s2_shelf_unblock_003 | 2 | shelf | shelf_unblock | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Slot bot_center is currently occupied by blue cube. First pick blue cube and place_on table to clear bot_center, then pick red cube, place_in_slot(bot_center), and done.</think><tool_call><function=pick><parameter=object>blue cube</parameter></function></tool_call><tool_call><function=place_on><parameter=target>... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "bot_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 219 | |||
s2_wh_nav_006_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s2_shelf_place_008_f | 2 | shelf | shelf_tag_or_rel | false | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think></think><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>bot_right</parameter></function></tool_call><tool_call><function=done><parameter=summary>Placed red cube in bot_right.</parameter></function></tool_call> | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", "mid_center", "mid_left"... | [] | 2.95 | 154 | |||
s2_wh_nav_012_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s1_tower_013_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate, then stack the red cube on the green cube, then stack the blue cube on the red cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \... | <think></think>[{"pick": "green cube"}, {"place_on": "yellow plate"}, {"pick": "red cube"}, {"place_on": "green cube"}, {"pick": "blue cube"}, {"place_on": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s1_tower_011_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the blue cube on the yellow plate, then stack the green cube on the blue cube, then stack the red cube on the green cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": ... | <think></think>[{"pick": "blue cube"}, {"place_on": "yellow plate"}, {"pick": "green cube"}, {"place_on": "blue cube"}, {"pick": "red cube"}, {"place_on": "green cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s3_shelf_unblock_013 | 3 | shelf | shelf_unblock | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Slot bot_center is currently occupied by blue cube. First pick blue cube and place_on table to clear bot_center, then pick green cube, place_in_slot(bot_center), and done.</think><tool_call><function=pick><parameter=object>blue cube</parameter></function></tool_call><tool_call><function=place_on><parameter=targe... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "bot_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 219 | |||
s2_shelf_place_004_t | 2 | shelf | shelf_tag_or_rel | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Looking at the shelf tags, the yellow tag marks the bot tier, so target slot is bot_right. Pick green cube from table, insert horizontally via place_in_slot(bot_right), then done.</think><tool_call><function=pick><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place_in_slot><p... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", "mid_center", "mid_left"... | [] | 3 | 195 | |||
s2_wh_nav_003_t | 2 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.7, -0.8), beyond R_max=0.55m from station 2. First dock_at_station('station 2'), pick('red cube'), then dock_at_station('station 1'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_cal... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 313 | |||
s1_unstack_011 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it,... | <think>The blue cube sits on top of the green cube. First pick blue cube and place on table to uncover green cube, then pick green cube onto yellow plate.</think>[{"pick": "blue cube"}, {"place_on": "table"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"blue cube": "green cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "... | [] | 2.9 | 176 | |||
s1_tower_003_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate, then stack the blue cube on the red cube, then stack the green cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"... | <think></think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "red cube"}, {"pick": "green cube"}, {"place_on": "blue cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s2_bi_swap_010_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently red cube sits on green cube on swap_zone, so green cube is blocked. 1) lift_and_hold(left, red cube) clears green cube; 2) pick(right, green cube); 3) place_under(left, green cube, swap_zone) restores red cube underneath; 4) place_on(right, red cube); 5) done.</think><tool_call><function=lift_and_hold>... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"red cube": "green cube", "green cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_... | [] | 3.1 | 364 | |||
s1_unstack_012 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it, b... | <think>The blue cube sits on top of the red cube. First pick blue cube and place on table to uncover red cube, then pick red cube onto yellow plate.</think>[{"pick": "blue cube"}, {"place_on": "table"}, {"pick": "red cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"blue cube": "red cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "st... | [] | 2.9 | 176 | |||
s2_shelf_place_010_f | 2 | shelf | shelf_tag_or_rel | false | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think></think><tool_call><function=pick><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>bot_center</parameter></function></tool_call><tool_call><function=done><parameter=summary>Placed green cube in bot_center.</parameter></function></tool_call> | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", "mid_center", "mid_left"... | [] | 2.95 | 155 | |||
s1_tower_006_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate, then stack the green cube on the red cube, then stack the blue cube on the green cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \... | <think></think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "green cube"}, {"place_on": "red cube"}, {"pick": "blue cube"}, {"place_on": "green cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s2_shelf_unblock_007 | 2 | shelf | shelf_unblock | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Slot bot_center is currently occupied by blue cube. First pick blue cube and place_on table to clear bot_center, then pick green cube, place_in_slot(bot_center), and done.</think><tool_call><function=pick><parameter=object>blue cube</parameter></function></tool_call><tool_call><function=place_on><parameter=targe... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "bot_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 219 | |||
s3_bi_swap_015_f | 3 | bimanual | bimanual_underneath_swap | false | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think></think><tool_call><function=lift_and_hold><parameter=arm>left</parameter><parameter=object>red cube</parameter><parameter=dz>0.18</parameter></function></tool_call><tool_call><function=pick><parameter=arm>right</parameter><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place_... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"red cube": "green cube", "green cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_... | [] | 2.95 | 284 | |||
s2_wh_nav_004_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s1_math_00 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "What is the remainder when 7^2026 is divided by 5? End with ANSWER: <number>."}] | <think>Powers of 7 mod 5 cycle with period 4 (2,4,3,1). 2026 mod 4 = 2, so 7^2 = 49 == 4 (mod 5).</think>ANSWER: 4 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 4 | 2.7 | 89 | ||
s1_tower_014_t | 1 | tabletop | dependency_tower | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate, then stack the blue cube on the green cube, then stack the red cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": ... | <think>All cubes (green cube, blue cube, red cube) and yellow plate sit separately on the table. Build bottom-up alternating pick and place_on: green cube->yellow plate, blue cube->green cube, red cube->blue cube.</think>[{"pick": "green cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "green c... | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.9 | 228 | |||
s2_wh_nav_000_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s1_point_012 | 1 | tabletop | pointing | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the green cube. Answer in JSON as a list of points: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}], with x and y as integers normalized to 0-1000."}]}] | <think></think>[{"point_2d": [311, 512], "label": "green cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [{"pred_norm": [311, 512], "gt_px": [198.95418876221845, 245.6834174564176], "mask_hit": true, "zone_ok": true}] | 2.85 | 83 | |||
s2_wh_nav_011_t | 2 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.6, -0.7), beyond R_max=0.55m from station 2. First dock_at_station('station 2'), pick('red cube'), then dock_at_station('station 1'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_cal... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 313 | |||
s3_shelf_place_014_t | 3 | shelf | shelf_tag_or_rel | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Looking at the shelf tags, the magenta tag marks the bot tier, so target slot is bot_center. Pick red cube from table, insert horizontally via place_in_slot(bot_center), then done.</think><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=place_in_slot><pa... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", "mid_center", "mid_left"... | [] | 3 | 197 | |||
s2_bi_handover_008 | 2 | bimanual | bimanual_handover | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Pick green cube with left arm, execute handover(left -> right), then place_on(right, swap_zone) and done.</think><tool_call><function=pick><parameter=arm>left</parameter><parameter=object>green cube</parameter></function></tool_call><tool_call><function=handover><parameter=from_arm>left</parameter><parameter=to_... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"blue cube": "red cube", "red cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy"... | [] | 3 | 255 | |||
s1_tower_004_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the blue cube on the yellow plate, then stack the red cube on the blue cube, then stack the green cube on the red cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"... | <think></think>[{"pick": "blue cube"}, {"place_on": "yellow plate"}, {"pick": "red cube"}, {"place_on": "blue cube"}, {"pick": "green cube"}, {"place_on": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s3_replan_005 | 3 | shelf | replan_recovery | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>The blue cube is already in mid_center, so do not re-move blue cube. Only green cube remains: pick(green cube), place_in_slot(top_center), done.</think><tool_call><function=pick><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>top_center</parameter... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "mid_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 218 | |||
s1_unstack_013 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it, b... | <think>The green cube sits on top of the red cube. First pick green cube and place on table to uncover red cube, then pick red cube onto yellow plate.</think>[{"pick": "green cube"}, {"place_on": "table"}, {"pick": "red cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"green cube": "red cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "s... | [] | 2.9 | 176 | |||
s2_bi_swap_012_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently red cube sits on green cube on swap_zone, so green cube is blocked. 1) lift_and_hold(left, red cube) clears green cube; 2) pick(right, green cube); 3) place_under(left, green cube, swap_zone) restores red cube underneath; 4) place_on(right, red cube); 5) done.</think><tool_call><function=lift_and_hold>... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"red cube": "green cube", "green cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_... | [] | 3.1 | 364 | |||
s2_shelf_pt_012 | 2 | shelf | pointing_shelf | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the red cube. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [561, 836], "label": "red cube"}] | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {},... | [{"pred_norm": [561, 836], "gt_px": [359.3241342930247, 401.25395690936205], "mask_hit": true, "zone_ok": true}] | 2.85 | 65 | |||
s1_excl_004 | 1 | tabletop | negative_exclusion | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put all cubes except the blue cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object... | <think>Visible cubes are blue cube, red cube, green cube. Exclude blue cube and yellow plate; move only red cube and green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": ["blue cube"], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_x... | [] | 3.05 | 178 | |||
s2_shelf_pt_014 | 2 | shelf | pointing_shelf | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the red cube. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [429, 715], "label": "red cube"}] | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {},... | [{"pred_norm": [429, 715], "gt_px": [274.38049948020824, 343.30240038362416], "mask_hit": true, "zone_ok": true}] | 2.85 | 65 | |||
s3_replan_002 | 3 | shelf | replan_recovery | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>The blue cube is already in mid_center, so do not re-move blue cube. Only red cube remains: pick(red cube), place_in_slot(top_center), done.</think><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>top_center</parameter></fun... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "mid_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 218 | |||
s1_excl_002 | 1 | tabletop | negative_exclusion | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put all cubes except the blue cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object... | <think>Visible cubes are blue cube, red cube, green cube. Exclude blue cube and yellow plate; move only red cube and green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": ["blue cube"], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_x... | [] | 3.05 | 178 | |||
s2_wh_nav_002_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s1_math_03 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "Pipe A fills in 6h, Pipe B fills in 3h, drain empties in 9h. Combined fill hours? End with ANSWER: <number>."}] | <think>Rate = 1/6 + 1/3 - 1/9 = 7/18 tank/hr, so fill time is 18/7 = 2.571 hours.</think>ANSWER: 18/7 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 18/7 | 2.7 | 96 | ||
s3_bi_handover_014 | 3 | bimanual | bimanual_handover | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Pick blue cube with left arm, execute handover(left -> right), then place_on(right, swap_zone) and done.</think><tool_call><function=pick><parameter=arm>left</parameter><parameter=object>blue cube</parameter></function></tool_call><tool_call><function=handover><parameter=from_arm>left</parameter><parameter=to_ar... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"green cube": "red cube", "red cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy... | [] | 3 | 255 | |||
s3_replan_000 | 3 | shelf | replan_recovery | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>The green cube is already in mid_center, so do not re-move green cube. Only blue cube remains: pick(blue cube), place_in_slot(top_center), done.</think><tool_call><function=pick><parameter=object>blue cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>top_center</parameter>... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"green cube": "mid_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right"... | [] | 3.1 | 218 | |||
s1_math_02 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "How many distinct real roots does x^4 - 5x^2 + 4 = 0 have? End with ANSWER: <number>."}] | <think>Factor (x^2 - 4)(x^2 - 1) = 0 -> x in {-2, -1, 1, 2}, giving 4 distinct real roots.</think>ANSWER: 4 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 4 | 2.7 | 89 | ||
s1_point_000 | 1 | tabletop | pointing | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue cube. Answer in JSON as a list of points: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}], with x and y as integers normalized to 0-1000."}]}] | <think></think>[{"point_2d": [660, 510], "label": "blue cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [{"pred_norm": [660, 510], "gt_px": [422.64401539628705, 244.84598676587598], "mask_hit": true, "zone_ok": true}] | 2.85 | 83 | |||
s1_tower_001_t | 1 | tabletop | dependency_tower | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate, then stack the blue cube on the red cube, then stack the green cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"... | <think>All cubes (red cube, blue cube, green cube) and yellow plate sit separately on the table. Build bottom-up alternating pick and place_on: red cube->yellow plate, blue cube->red cube, green cube->blue cube.</think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "red cube"},... | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.9 | 228 | |||
s2_shelf_unblock_005 | 2 | shelf | shelf_unblock | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>Slot top_center is currently occupied by red cube. First pick red cube and place_on table to clear top_center, then pick blue cube, place_in_slot(top_center), and done.</think><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=place_on><parameter=target>ta... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"red cube": "top_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right", ... | [] | 3.1 | 218 | |||
s1_math_08 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "What is the remainder when 7^2026 is divided by 5? End with ANSWER: <number>."}] | <think>Powers of 7 mod 5 cycle with period 4 (2,4,3,1). 2026 mod 4 = 2, so 7^2 = 49 == 4 (mod 5).</think>ANSWER: 4 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 4 | 2.7 | 89 | ||
s2_wh_nav_003_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>station 1</parameter></function></tool_call><tool_... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 251 | |||
s1_point_005 | 1 | tabletop | pointing | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the red cube. Answer in JSON as a list of points: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}], with x and y as integers normalized to 0-1000."}]}] | <think></think>[{"point_2d": [689, 487], "label": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [{"pred_norm": [689, 487], "gt_px": [440.7352211875564, 233.76080445212793], "mask_hit": true, "zone_ok": true}] | 2.85 | 83 | |||
s2_wh_nav_000_t | 2 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.7, -0.6), beyond R_max=0.55m from station 2. First dock_at_station('blue zone'), pick('red cube'), then dock_at_station('red zone'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 314 | |||
s2_wh_pt_010 | 2 | wheeled | pointing_nav_pad | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue docking pad at station 2. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [590, 301], "label": "blue docking pad"}] | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station":... | [{"pred_norm": [590, 301], "gt_px": [377.3050463439158, 144.49158942680697], "mask_hit": true, "zone_ok": true}] | 2.85 | 73 | |||
s1_point_013 | 1 | tabletop | pointing | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the red cube. Answer in JSON as a list of points: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}], with x and y as integers normalized to 0-1000."}]}] | <think></think>[{"point_2d": [374, 407], "label": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [{"pred_norm": [374, 407], "gt_px": [239.56179598995598, 195.27815857196134], "mask_hit": true, "zone_ok": true}] | 2.85 | 83 | |||
s1_excl_014 | 1 | tabletop | negative_exclusion | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put all cubes except the green cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one objec... | <think>Visible cubes are green cube, blue cube, red cube. Exclude green cube and yellow plate; move only blue cube and red cube onto yellow plate.</think>[{"pick": "blue cube"}, {"place_on": "yellow plate"}, {"pick": "red cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": ["green cube"], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_... | [] | 3.05 | 178 | |||
s1_unstack_009 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the blue cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it, ... | <think>The green cube sits on top of the blue cube. First pick green cube and place on table to uncover blue cube, then pick blue cube onto yellow plate.</think>[{"pick": "green cube"}, {"place_on": "table"}, {"pick": "blue cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"green cube": "blue cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "... | [] | 2.9 | 176 | |||
s1_tower_009_t | 1 | tabletop | dependency_tower | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate, then stack the blue cube on the green cube, then stack the red cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": ... | <think>All cubes (green cube, blue cube, red cube) and yellow plate sit separately on the table. Build bottom-up alternating pick and place_on: green cube->yellow plate, blue cube->green cube, red cube->blue cube.</think>[{"pick": "green cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "green c... | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.9 | 228 | |||
s2_bi_swap_003_f | 2 | bimanual | bimanual_underneath_swap | false | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think></think><tool_call><function=lift_and_hold><parameter=arm>left</parameter><parameter=object>blue cube</parameter><parameter=dz>0.18</parameter></function></tool_call><tool_call><function=pick><parameter=arm>right</parameter><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"blue cube": "green cube", "green cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot... | [] | 2.95 | 284 | |||
s1_tower_007_t | 1 | tabletop | dependency_tower | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the red cube on the yellow plate, then stack the blue cube on the red cube, then stack the green cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"... | <think>All cubes (red cube, blue cube, green cube) and yellow plate sit separately on the table. Build bottom-up alternating pick and place_on: red cube->yellow plate, blue cube->red cube, green cube->blue cube.</think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "red cube"},... | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.9 | 228 | |||
s2_wh_nav_002_t | 2 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.8, -0.7), beyond R_max=0.55m from station 2. First dock_at_station('blue zone'), pick('red cube'), then dock_at_station('red zone'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 314 | |||
s3_wh_nav_015_t | 3 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.6, -0.8), beyond R_max=0.55m from station 2. First dock_at_station('station 2'), pick('red cube'), then dock_at_station('station 1'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_cal... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 313 | |||
s3_replan_001 | 3 | shelf | replan_recovery | true | 2 | [{"role": "system", "content": "You control a 5-DOF robot arm facing a 3-tier x 3-slot shelf and table. Tools: pick(object), place_in_slot(slot), place_on(target), done(summary). Slots: {top,mid,bot}_{left,center,right}. Output ONLY valid <tool_call> blocks."}, {"role": "user", "content": [{"type": "image"}, {"type": "... | <think>The blue cube is already in mid_center, so do not re-move blue cube. Only green cube remains: pick(green cube), place_in_slot(top_center), done.</think><tool_call><function=pick><parameter=object>green cube</parameter></function></tool_call><tool_call><function=place_in_slot><parameter=slot>top_center</parameter... | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {"blue cube": "mid_center"}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": ["bot_center", "bot_left", "bot_right",... | [] | 3.1 | 218 | |||
s1_unstack_006 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it,... | <think>The red cube sits on top of the green cube. First pick red cube and place on table to uncover green cube, then pick green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "table"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"red cube": "green cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "s... | [] | 2.9 | 176 | |||
s3_wh_nav_016_f | 3 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>blue zone</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>red zone</parameter></function></tool_call><tool_c... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 254 | |||
s1_tower_000_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the blue cube on the yellow plate, then stack the red cube on the blue cube, then stack the green cube on the red cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"... | <think></think>[{"pick": "blue cube"}, {"place_on": "yellow plate"}, {"pick": "red cube"}, {"place_on": "blue cube"}, {"pick": "green cube"}, {"place_on": "red cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 | |||
s2_shelf_pt_000 | 2 | shelf | pointing_shelf | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue cube. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [536, 875], "label": "blue cube"}] | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {},... | [{"pred_norm": [536, 875], "gt_px": [342.8291950962119, 419.9072427045481], "mask_hit": true, "zone_ok": true}] | 2.85 | 65 | |||
s2_wh_nav_007_f | 2 | wheeled | wheeled_mobile_manip | false | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think></think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_call><tool_call><function=pick><parameter=object>red cube</parameter></function></tool_call><tool_call><function=dock_at_station><parameter=station_label>station 1</parameter></function></tool_call><tool_... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 2.95 | 251 | |||
s3_bi_swap_013_t | 3 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently green cube sits on red cube on swap_zone, so red cube is blocked. 1) lift_and_hold(left, green cube) clears red cube; 2) pick(right, red cube); 3) place_under(left, red cube, swap_zone) restores green cube underneath; 4) place_on(right, green cube); 5) done.</think><tool_call><function=lift_and_hold><p... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"green cube": "red cube", "red cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy... | [] | 3.1 | 364 | |||
s2_shelf_pt_002 | 2 | shelf | pointing_shelf | false | 1 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "Point to the blue cube. Answer in JSON: [{\"point_2d\": [x, y], \"label\": \"<short label>\"}]."}]}] | <think></think>[{"point_2d": [351, 648], "label": "blue cube"}] | {"embodiment": "shelf", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {},... | [{"pred_norm": [351, 648], "gt_px": [224.32795392860288, 310.92588560688637], "mask_hit": true, "zone_ok": true}] | 2.85 | 65 | |||
s2_wh_nav_007_t | 2 | wheeled | wheeled_mobile_manip | true | 3 | [{"role": "system", "content": "You control a 3-DOF wheeled robot carrying a 4-DOF arm across station 1 (red zone), station 2 (blue zone), and station 3 (green zone). Tools: dock_at_station(station_label), drive_to_point(point_2d, camera, stop_dist_m), move_base(forward_m, turn_deg), pick(object), place_on(target), don... | <think>The mobile base starts near (-0.8, -0.8), beyond R_max=0.55m from station 2. First dock_at_station('station 2'), pick('red cube'), then dock_at_station('station 1'), place_on('yellow plate'), and done.</think><tool_call><function=dock_at_station><parameter=station_label>station 2</parameter></function></tool_cal... | {"embodiment": "wheeled", "scene_objects": ["blue cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {"red cube": [0.85, 0.75], "yellow plate": [0.83, -0.77]}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "... | [] | 3.1 | 313 | |||
s1_excl_000 | 1 | tabletop | negative_exclusion | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put all cubes except the blue cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object... | <think>Visible cubes are blue cube, red cube, green cube. Exclude blue cube and yellow plate; move only red cube and green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "yellow plate"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": ["blue cube"], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_x... | [] | 3.05 | 178 | |||
s2_bi_swap_000_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently green cube sits on red cube on swap_zone, so red cube is blocked. 1) lift_and_hold(left, green cube) clears red cube; 2) pick(right, red cube); 3) place_under(left, red cube, swap_zone) restores green cube underneath; 4) place_on(right, green cube); 5) done.</think><tool_call><function=lift_and_hold><p... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"green cube": "red cube", "red cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy... | [] | 3.1 | 364 | |||
s2_bi_handover_011 | 2 | bimanual | bimanual_handover | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Pick green cube with left arm, execute handover(left -> right), then place_on(right, swap_zone) and done.</think><tool_call><function=pick><parameter=arm>left</parameter><parameter=object>green cube</parameter></function></tool_call><tool_call><function=handover><parameter=from_arm>left</parameter><parameter=to_... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"red cube": "blue cube", "blue cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy... | [] | 3 | 255 | |||
s1_math_05 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "A 2 kg cart at 3 m/s collides and sticks to a stationary 1 kg cart. Final speed in m/s? End with ANSWER: <number>."}] | <think>Conservation of momentum: p = 2*3 = 6 kg*m/s. Total mass = 3 kg, so v = 6/3 = 2 m/s.</think>ANSWER: 2 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 2 | 2.7 | 91 | ||
s2_bi_swap_003_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently blue cube sits on green cube on swap_zone, so green cube is blocked. 1) lift_and_hold(left, blue cube) clears green cube; 2) pick(right, green cube); 3) place_under(left, green cube, swap_zone) restores blue cube underneath; 4) place_on(right, blue cube); 5) done.</think><tool_call><function=lift_and_h... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"blue cube": "green cube", "green cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot... | [] | 3.1 | 364 | |||
s1_math_09 | 1 | text_reasoning | concise_math_budget | true | 1 | [{"role": "user", "content": "A 2 kg cart at 3 m/s collides and sticks to a stationary 1 kg cart. Final speed in m/s? End with ANSWER: <number>."}] | <think>Conservation of momentum: p = 2*3 = 6 kg*m/s. Total mass = 3 kg, so v = 6/3 = 2 m/s.</think>ANSWER: 2 | {"embodiment": "text", "scene_objects": [], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zone_to_station": {}, "obstacles_xy": [], "base_xy": null, ... | [] | ANSWER: 2 | 2.7 | 91 | ||
s2_bi_swap_006_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently blue cube sits on red cube on swap_zone, so red cube is blocked. 1) lift_and_hold(left, blue cube) clears red cube; 2) pick(right, red cube); 3) place_under(left, red cube, swap_zone) restores blue cube underneath; 4) place_on(right, blue cube); 5) done.</think><tool_call><function=lift_and_hold><param... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"blue cube": "red cube", "red cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy"... | [] | 3.1 | 364 | |||
s2_bi_swap_004_t | 2 | bimanual | bimanual_underneath_swap | true | 3 | [{"role": "system", "content": "You control two 5-DOF robot arms (left and right) over a shared swap_zone and staging_pad. Tools: lift_and_hold(arm, object, dz), pick(arm, object), place_under(arm, target_above, base_surface), place_on(arm, target), handover(from_arm, to_arm), done(summary). Never pick a buried object ... | <think>Currently green cube sits on blue cube on swap_zone, so blue cube is blocked. 1) lift_and_hold(left, green cube) clears blue cube; 2) pick(right, blue cube); 3) place_under(left, blue cube, swap_zone) restores green cube underneath; 4) place_on(right, green cube); 5) done.</think><tool_call><function=lift_and_ho... | {"embodiment": "bimanual", "scene_objects": ["blue cube", "green cube", "red cube"], "excluded_objects": [], "on_top": {"green cube": "blue cube", "blue cube": "swap_zone"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_... | [] | 3.1 | 364 | |||
s1_unstack_005 | 1 | tabletop | unstack_blocker | true | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": \"<object>\"} or {\"place_on\": \"<object>\"}. Pick exactly one object, then place it,... | <think>The red cube sits on top of the green cube. First pick red cube and place on table to uncover green cube, then pick green cube onto yellow plate.</think>[{"pick": "red cube"}, {"place_on": "table"}, {"pick": "green cube"}, {"place_on": "yellow plate"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {"red cube": "green cube"}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "s... | [] | 2.9 | 176 | |||
s1_tower_009_f | 1 | tabletop | dependency_tower | false | 2 | [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": "You control a robot arm above this table. Task: put the green cube on the yellow plate, then stack the blue cube on the green cube, then stack the red cube on the blue cube\nWrite the robot's steps as a JSON list. Each step is either {\"pick\": ... | <think></think>[{"pick": "green cube"}, {"place_on": "yellow plate"}, {"pick": "blue cube"}, {"place_on": "green cube"}, {"pick": "red cube"}, {"place_on": "blue cube"}] | {"embodiment": "tabletop", "scene_objects": ["blue cube", "green cube", "red cube", "yellow plate"], "excluded_objects": [], "on_top": {}, "in_slot": {}, "object_xy": {}, "valid_surfaces": ["staging_pad", "swap_zone", "table", "yellow plate", "yellow_plate"], "valid_slots": [], "slot_xy": {}, "station_dock_xy": {}, "zo... | [] | 2.75 | 180 |
Shrey160/lfm3-vl-rl-v1)
Verified Reinforcement Learning with Verifiable Rewards (RLVR — Dr.GRPO + DAPO) dataset and Tier-1 StateBank built in MuJoCo 4.0 for LiquidAI/LFM3-VL-2B across tabletop, shelf manipulation, bimanual swaps, and wheeled mobile manipulation.
207 (train: 186, validation: 21)84 | Stage 2: 86 | Stage 3: 37think=True (<think>...</think>): 118 | think=False (<think></think>): 89100% LFM3-VL-2B chat_template.jinja + SymbolicWorldSim verified (mean oracle R_total = 2.942)| Embodiment | Rows | Share |
|---|---|---|
bimanual |
30 | 14.5% |
shelf |
48 | 23.2% |
tabletop |
72 | 34.8% |
text_reasoning |
12 | 5.8% |
wheeled |
45 | 21.7% |
data/train-00000-of-00001.parquet, data/validation-00000-of-00001.parquet: HF Parquet splits with embedded PIL RGB imagestrain.jsonl, validation.jsonl, state_bank.jsonl: full RLVR prompt + SymbolicWorldState JSONL recordsid: unique sample identifierstage: curriculum stage (1 = stabilization/tabletop, 2 = multi-embodiment symbolic RL, 3 = physics/replanning)embodiment: tabletop, shelf, bimanual, wheeled, or text_reasoningcategory: fine-grained task typethink_mode: boolean (True for bounded <think>...</think>, False for direct <think></think>)difficulty_tier: integer difficulty tier (1..3) controlling piecewise thinking-budget target L_target(d)prompt_messages_json: JSON-encoded LFM3 ChatML system + multimodal user prompt ({"type": "image"} slots)tools_json: optional OpenAI tools JSON string ("" when inline <tool_call> prompt schema is used)images: Sequence(Image()) (1–2 camera RGB views or [] for text-only anchors)oracle_completion: reference verified <think>...</think><tool_call>...</tool_call> completionworld_state_json: serialized SymbolicWorldState for Tier-1 < 1 ms SymbolicWorldSim rollout verificationpoint_predictions_json: optional normalized [0..1000] point_2d grounding metadataexpected_answer: optional exact answer string for Stage 1 math/IF anchorsoracle_r_total: verified 6-component RLVR oracle rewardn_tokens: verified ChatML prompt+completion token lengthimport json
from datasets import load_dataset
ds = load_dataset("Shrey160/lfm3-vl-rl-v1")
row = ds["train"][0]
messages = json.loads(row["prompt_messages_json"])
world_state = json.loads(row["world_state_json"])
images = row["images"]