Datasets:

dexsuite
/

spray_graffiti_floating_sharpa_scripted

@@ -31,7 +31,7 @@ Scripted-expert DexSuite data for **spray_graffiti** with the floating wrist and
 The controller grasps the can with a wrap grasp along one of ten approach routes, carries it over the paper, aligns the nozzle with the measured paint hit and the index fingertip with the button, and paints overlapping passes inside the stencil. It keeps painting after the first success (to about 90% coverage) and then releases the button, so `success` and `terminated` are true on every frame from the first successful transition on (the last 102 frames of an episode at the median). DART: during recording the environment executed the controller's command plus a smooth Ornstein-Uhlenbeck perturbation of the wrist pose (1.5 cm and 4 degrees, 0.6 s time constant, scaled by 1.0 / 0.5 / 0.25 in the free / approach / in-hand phases and switched off while the nozzle and the index pad are aligned), and the recorded action is the controller's clean command, its correction from the perturbed state (Laskey et al., 2017). The command the environment actually executed at step t is stored in `obs.state.other.action` of frame t+1, so replaying `action` open loop does not reproduce the recorded episodes. Each episode randomizes the can and paper positions (±2 cm in x, ±3 cm in y), the can yaw (±15°), the start wrist pose (±5 cm and ±12° per axis) and the relaxed finger pose; 200 of 236 attempts were kept, the other 36 failed the controller's own checks (mostly a grip not secured during the approach) and were discarded.
-The robot is an arm-less floating wrist (a virtual 6-DoF gantry) carrying the right Sharpa hand (22 DoF); the 29-D action is 7 `osc_pose_abs_quat` values (position offset and orientation relative to the episode-start wrist pose) followed by 22 hand joint position targets in radians. The recording names the hand `sharpa_fardexbot`, the Sharpa hand with its wrist-camera mount, which shows in the front-camera frames; DexSuite registers that name as an alias of the plain `sharpa` hand. Besides the standard DexSuite keys, every frame carries per-finger contact forces (`obs.state.other.finger_force_world`, `finger_force_norm` in N, `finger_force_display`, `finger_force_saturated`, `finger_contact`; thumb, index, middle, ring, pinky) and the controller's phase (`obs.state.other.controller_phase`: 0 free, 1 approach, 2 in hand) and stage (`obs.state.other.controller_stage`, named in `meta/controller.json`). `meta/env_config.json` rebuilds the environment and `meta/controller_episodes/episode_*.json` keeps each episode's scene seed, style seed and controller parameters. Depth was rendered during collection but is not included.
 - **Homepage:** https://huggingface.co/dexsuite
 - **Paper:** https://arxiv.org/abs/2610.37799

 The controller grasps the can with a wrap grasp along one of ten approach routes, carries it over the paper, aligns the nozzle with the measured paint hit and the index fingertip with the button, and paints overlapping passes inside the stencil. It keeps painting after the first success (to about 90% coverage) and then releases the button, so `success` and `terminated` are true on every frame from the first successful transition on (the last 102 frames of an episode at the median). DART: during recording the environment executed the controller's command plus a smooth Ornstein-Uhlenbeck perturbation of the wrist pose (1.5 cm and 4 degrees, 0.6 s time constant, scaled by 1.0 / 0.5 / 0.25 in the free / approach / in-hand phases and switched off while the nozzle and the index pad are aligned), and the recorded action is the controller's clean command, its correction from the perturbed state (Laskey et al., 2017). The command the environment actually executed at step t is stored in `obs.state.other.action` of frame t+1, so replaying `action` open loop does not reproduce the recorded episodes. Each episode randomizes the can and paper positions (±2 cm in x, ±3 cm in y), the can yaw (±15°), the start wrist pose (±5 cm and ±12° per axis) and the relaxed finger pose; 200 of 236 attempts were kept, the other 36 failed the controller's own checks (mostly a grip not secured during the approach) and were discarded.
+The robot is an arm-less floating wrist (a virtual 6-DoF gantry) carrying the right Sharpa hand (22 DoF); the 29-D action is 7 `osc_pose_abs_quat` values (position offset and orientation relative to the episode-start wrist pose) followed by 22 hand joint position targets in radians. The hand is `sharpa_fardexbot`, the right Sharpa hand carrying the wrist-camera mount of the real handheld rig (mount plate, quick changer and D405 body, which collide with the scene and show in the front-camera frames); DexSuite ships it under that name. Besides the standard DexSuite keys, every frame carries per-finger contact forces (`obs.state.other.finger_force_world`, `finger_force_norm` in N, `finger_force_display`, `finger_force_saturated`, `finger_contact`; thumb, index, middle, ring, pinky) and the controller's phase (`obs.state.other.controller_phase`: 0 free, 1 approach, 2 in hand) and stage (`obs.state.other.controller_stage`, named in `meta/controller.json`). `meta/env_config.json` rebuilds the environment and `meta/controller_episodes/episode_*.json` keeps each episode's scene seed, style seed and controller parameters. Depth was rendered during collection but is not included.
 - **Homepage:** https://huggingface.co/dexsuite
 - **Paper:** https://arxiv.org/abs/2610.37799