Apppublic

connect-ai-lab-aimo/physical-ai-1

sourceHugging Faceupdated 29d agoView on Hugging Face

0likes

App README

🦾 LeRobot Physical AI Simulator (Static Space)

Hugging Face LeRobot & PyTorch Embodied Control Engine | Connect AI LAB Physical AI & T-Shape Alignment Demonstration Data Collection in Pure JavaScript

📌 Overview

This Space is a 100% client-side, browser-based T-Shape Object Alignment Teleoperation Simulator designed for Physical AI and Imitation Learning (ACT / Diffusion Policy / LeRobot) enthusiasts.

Manipulate the Gray Movable T using mouse drag and rotation controls to align it precisely with the Green Target T silhouette. As you perform teleoperation, your demonstration trajectory $(X, Y, \theta, \Delta X, \Delta Y, \Delta\theta, \text{Error}, \text{Reward})$ is recorded and can be exported directly as JSON and CSV datasets.

🎮 Controls

Action	Function	Details
Left Mouse Drag	Move T-Object $(X, Y)$	Click and drag the gray T to move it across the canvas
Mouse Wheel	Rotate T-Object ($\pm 5^\circ$)	Scroll up/down to rotate
Right Mouse Drag	Free Angle Rotation	Right-click and drag in the desired direction
Blue Handle Drag	Rotation Knob	Drag the top blue knob to orient the T-block
`Q` / `E` (or `A` / `D`)	Fine Rotation	Rotate counter-clockwise / clockwise in 5° increments
`S` Key / Button	Start / Stop Recording	Record demonstration frames
`N` Key / Button	Next Episode	Save current buffer and randomize target pose
`R` Key / Button	Reset Target	Re-randomize target pose
`Space` Key / Button	Pause / Resume	Pause or resume simulation

💾 Dataset Export & Local Storage

—100% Client-Side Privacy: No demonstration data is uploaded to external servers.
—IndexedDB Persistence: Recorded episodes are automatically cached in browser storage across page reloads.
—Direct Downloads:
—📥 Episode JSON / 📥 Episode CSV: Download the latest completed episode.
—📦 All Episodes (JSON) / 📦 All Episodes (CSV): Export the entire multi-episode dataset in one file.

📊 Recorded Trajectory Schema

Every step logs:

json

{
  "timestamp": 1787600980.08,
  "episode_id": 1,
  "step": 1,
  "object_x": 231.92,
  "object_y": 325.68,
  "object_rotation": 38.84,
  "previous_x": 231.92,
  "previous_y": 325.68,
  "previous_rotation": 38.84,
  "delta_x": 0.0,
  "delta_y": 0.0,
  "delta_rotation": 0.0,
  "target_x": 266.83,
  "target_y": 471.03,
  "target_rotation": 291.95,
  "position_error": 149.49,
  "rotation_error": 106.89,
  "goal_alignment": 52.2,
  "reward": 0.1323,
  "success": false,
  "mouse_action": "MOUSE_DRAG_MOVE"
}