Datasets:
Download documents/interfaces/interface-76b013ef.json from Agent-as-Policy/agent-as-policy-sim: direct link, hf CLI and curl.
- Browser
- Download file 12.6 kB
-
https://huggingface.co/datasets/Agent-as-Policy/agent-as-policy-sim/resolve/main/documents/interfaces/interface-76b013ef.json
- Command line
-
hf download hf://datasets/Agent-as-Policy/agent-as-policy-sim/documents/interfaces/interface-76b013ef.json
-
curl -L -o interface-76b013ef.json https://huggingface.co/datasets/Agent-as-Policy/agent-as-policy-sim/resolve/main/documents/interfaces/interface-76b013ef.json
12.6 kB
| { | |
| "server": "astra_control", | |
| "instructions": "Each observe/step result starts with one compact JSON text block, followed by labelled top, left_wrist and right_wrist images. Never retry an uncertain action with a new action_id; use action_status.", | |
| "tools": [ | |
| { | |
| "name": "goal", | |
| "description": "Return the task instruction.", | |
| "inputSchema": { | |
| "properties": {}, | |
| "title": "goalArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "observe", | |
| "description": "Current observation: compact state of both arms and top, left_wrist, right_wrist images. No physics advances.", | |
| "inputSchema": { | |
| "properties": {}, | |
| "title": "observeArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "measure", | |
| "description": "3D point at pixel (u,v) of one camera in the latest observation, expressed in the named arm's base frame (m). The images have no depth: with plane_z at the table (<= 0.005) the pixel ray is intersected with that horizontal plane; otherwise the height comes from matching the pixel in a second camera (top <-> that arm's wrist), returned as z_from match or uncertain with a review image, and a higher plane_z is only the fallback, used when the match is uncertain, missing or more than 0.02 m away from it (z_from fallback). on_robot marks a pixel inside a robot arm's outline. Returns measurement ids for step evidence. A query may give cands [[u,v],...] instead of u,v: all are measured and one is kept, the most reliable (pick solid: a plane or confident match before a fallback or uncertain one, off the robot) or the highest/lowest z (pick high|low), after dropping points outside zr [zmin,zmax]. label names the kept point for later run steps. A query may also be a string '[label:] L|R top|lw|rw u,v[/u,v...] [z=..] [pick=..] [zr=a,b] target text'. pose: true (string option `pose`, top camera only) also returns the footprint of the object at that pixel seen from above: xy centroid, axis_deg / tail_deg heading, edge_deg, length_mm, width_mm, status.", | |
| "inputSchema": { | |
| "$defs": { | |
| "PosePointQuery": { | |
| "additionalProperties": false, | |
| "description": "A measure query at one pixel; pose: true (top camera only) also returns the footprint of the object seen from above.", | |
| "properties": { | |
| "arm": { | |
| "enum": [ | |
| "left", | |
| "right" | |
| ], | |
| "title": "Arm", | |
| "type": "string" | |
| }, | |
| "cam": { | |
| "enum": [ | |
| "top", | |
| "left_wrist", | |
| "right_wrist" | |
| ], | |
| "title": "Cam", | |
| "type": "string" | |
| }, | |
| "u": { | |
| "anyOf": [ | |
| { | |
| "maximum": 639, | |
| "minimum": 0, | |
| "type": "integer" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "U" | |
| }, | |
| "v": { | |
| "anyOf": [ | |
| { | |
| "maximum": 479, | |
| "minimum": 0, | |
| "type": "integer" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "V" | |
| }, | |
| "target": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Target", | |
| "type": "string" | |
| }, | |
| "visibility": { | |
| "anyOf": [ | |
| { | |
| "maxLength": 200, | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Visibility" | |
| }, | |
| "plane_z": { | |
| "anyOf": [ | |
| { | |
| "type": "number" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Plane Z" | |
| }, | |
| "height_basis": { | |
| "anyOf": [ | |
| { | |
| "maxLength": 200, | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Height Basis" | |
| }, | |
| "cands": { | |
| "anyOf": [ | |
| { | |
| "items": { | |
| "items": { | |
| "type": "integer" | |
| }, | |
| "maxItems": 2, | |
| "minItems": 2, | |
| "type": "array" | |
| }, | |
| "maxItems": 8, | |
| "minItems": 1, | |
| "type": "array" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Cands" | |
| }, | |
| "pick": { | |
| "anyOf": [ | |
| { | |
| "enum": [ | |
| "solid", | |
| "high", | |
| "low" | |
| ], | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Pick" | |
| }, | |
| "zr": { | |
| "anyOf": [ | |
| { | |
| "items": { | |
| "type": "number" | |
| }, | |
| "maxItems": 2, | |
| "minItems": 2, | |
| "type": "array" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Zr" | |
| }, | |
| "label": { | |
| "anyOf": [ | |
| { | |
| "maxLength": 40, | |
| "minLength": 1, | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Label" | |
| }, | |
| "pose": { | |
| "anyOf": [ | |
| { | |
| "type": "boolean" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Pose" | |
| } | |
| }, | |
| "required": [ | |
| "arm", | |
| "cam", | |
| "target" | |
| ], | |
| "title": "PosePointQuery", | |
| "type": "object" | |
| } | |
| }, | |
| "properties": { | |
| "observation_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Observation Id", | |
| "type": "string" | |
| }, | |
| "queries": { | |
| "items": { | |
| "anyOf": [ | |
| { | |
| "$ref": "#/$defs/PosePointQuery" | |
| }, | |
| { | |
| "type": "string" | |
| } | |
| ] | |
| }, | |
| "maxItems": 8, | |
| "minItems": 1, | |
| "title": "Queries", | |
| "type": "array" | |
| } | |
| }, | |
| "required": [ | |
| "observation_id", | |
| "queries" | |
| ], | |
| "title": "measureArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "step", | |
| "description": "Execute one transaction for one or both arms on a shared clock; return compact outcome and fresh images. actions maps left/right to {command, args, intent, measurement_id?, evidence?}. Commands: move_ee(position{x,y,z} or position_offset + measurement_id, rotation{w,x,y,z}, mode plan|linear, speed_profile precision|transit), move_delta(dpos[3] m, optional drot_deg[3], speed_profile), move_joints(joints[6]), home, gripper(action open|close, or action width with width_mm: the pad gap in mm, 0.8 closed .. 88.8 fully open), hold(samples 1..25). Motion options: max_samples, settle_samples 0..10, position_tolerance_m, rotation_tolerance_rad.", | |
| "inputSchema": { | |
| "$defs": { | |
| "ArmAction": { | |
| "additionalProperties": false, | |
| "properties": { | |
| "command": { | |
| "enum": [ | |
| "move_ee", | |
| "move_delta", | |
| "move_joints", | |
| "home", | |
| "gripper", | |
| "hold" | |
| ], | |
| "title": "Command", | |
| "type": "string" | |
| }, | |
| "args": { | |
| "additionalProperties": true, | |
| "title": "Args", | |
| "type": "object" | |
| }, | |
| "intent": { | |
| "enum": [ | |
| "observe", | |
| "grasp", | |
| "transfer", | |
| "contact", | |
| "retreat", | |
| "place", | |
| "manipulate" | |
| ], | |
| "title": "Intent", | |
| "type": "string" | |
| }, | |
| "measurement_id": { | |
| "anyOf": [ | |
| { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Measurement Id" | |
| }, | |
| "evidence": { | |
| "anyOf": [ | |
| { | |
| "additionalProperties": true, | |
| "type": "object" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Evidence" | |
| } | |
| }, | |
| "required": [ | |
| "command", | |
| "args", | |
| "intent" | |
| ], | |
| "title": "ArmAction", | |
| "type": "object" | |
| } | |
| }, | |
| "properties": { | |
| "action_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Action Id", | |
| "type": "string" | |
| }, | |
| "observation_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Observation Id", | |
| "type": "string" | |
| }, | |
| "actions": { | |
| "additionalProperties": { | |
| "$ref": "#/$defs/ArmAction" | |
| }, | |
| "maxProperties": 2, | |
| "minProperties": 1, | |
| "propertyNames": { | |
| "enum": [ | |
| "left", | |
| "right" | |
| ] | |
| }, | |
| "title": "Actions", | |
| "type": "object" | |
| } | |
| }, | |
| "required": [ | |
| "action_id", | |
| "observation_id", | |
| "actions" | |
| ], | |
| "title": "stepArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "run", | |
| "description": "Execute up to 20 steps in order, server-side, stopping at the first surprise. Ids and observation threading are automatic; every action still passes the normal runtime checks. Step = {left?: arm, right?: arm, expect?: {arm: {gap_min_mm, gap_max_mm}}} or {measure: [queries], as: [labels]}. Arm = {do: ee|delta|open|close|home|hold|joints, intent?, to: [x,y,z] | {at: label_or_measurement_id, d: [dx,dy,dz]}, rot?: {w,x,y,z} | {down: deg} | {tilted: [az_deg, tilt_deg]} (default: keep current), mode?: linear|plan, speed?: precision|transit (default transit for observe/transfer/retreat intents), d: [dx,dy,dz] for delta, m?: measurement label or id, ev?: {...}, samples?}. show = {at: {\"<k>\": [labels]}, end: [labels], on_stop: [labels]}; evidence is shared by every arm action. A step may instead be a command string, e.g. 'L ee g+0,0,.05 down90 plan obs & R d 0,0,.03 retreat', 'B close gap>2', 'L open support=flat surface below', or a measure step '? g: L top 180,214 near object top'. show may be a string such as 'top lw 2:rw stop:top'. Retreat moves and home get a default clearance when none is given. do: width with width_mm (the pad gap in mm, 0.8 closed .. 88.8 fully open) moves the fingers to that gap, e.g. 'L width 30'. A measure query may add `pose` (e.g. '? t: L top 180,214 z=.015 pose object top') for the object's footprint.", | |
| "inputSchema": { | |
| "properties": { | |
| "observation_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Observation Id", | |
| "type": "string" | |
| }, | |
| "steps": { | |
| "items": { | |
| "anyOf": [ | |
| { | |
| "additionalProperties": true, | |
| "type": "object" | |
| }, | |
| { | |
| "type": "string" | |
| } | |
| ] | |
| }, | |
| "maxItems": 20, | |
| "minItems": 1, | |
| "title": "Steps", | |
| "type": "array" | |
| }, | |
| "show": { | |
| "anyOf": [ | |
| { | |
| "additionalProperties": true, | |
| "type": "object" | |
| }, | |
| { | |
| "items": { | |
| "type": "string" | |
| }, | |
| "type": "array" | |
| }, | |
| { | |
| "type": "string" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Show" | |
| }, | |
| "evidence": { | |
| "anyOf": [ | |
| { | |
| "additionalProperties": true, | |
| "type": "object" | |
| }, | |
| { | |
| "type": "null" | |
| } | |
| ], | |
| "default": null, | |
| "title": "Evidence" | |
| } | |
| }, | |
| "required": [ | |
| "observation_id", | |
| "steps" | |
| ], | |
| "title": "runArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "action_status", | |
| "description": "Reconcile an existing action after an uncertain result. Never dispatches motion.", | |
| "inputSchema": { | |
| "properties": { | |
| "action_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Action Id", | |
| "type": "string" | |
| } | |
| }, | |
| "required": [ | |
| "action_id" | |
| ], | |
| "title": "action_statusArguments", | |
| "type": "object" | |
| } | |
| }, | |
| { | |
| "name": "assess", | |
| "description": "Record your final task judgement with evidence.summary. The benchmark evaluator decides independently.", | |
| "inputSchema": { | |
| "properties": { | |
| "observation_id": { | |
| "maxLength": 160, | |
| "minLength": 1, | |
| "title": "Observation Id", | |
| "type": "string" | |
| }, | |
| "outcome": { | |
| "enum": [ | |
| "success", | |
| "failure", | |
| "unknown" | |
| ], | |
| "title": "Outcome", | |
| "type": "string" | |
| }, | |
| "evidence": { | |
| "additionalProperties": true, | |
| "title": "Evidence", | |
| "type": "object" | |
| } | |
| }, | |
| "required": [ | |
| "observation_id", | |
| "outcome", | |
| "evidence" | |
| ], | |
| "title": "assessArguments", | |
| "type": "object" | |
| } | |
| } | |
| ] | |
| } | |