🦾
Robotics Diffusion Policy Studio v1.0 Live

Embodied AI, Visuomotor DDIM Control & 3-DOF Inverse Kinematics

Receding-Horizon Visuomotor Control

Policy Architecture & Denoising Controller

Active Policy:
DDIM Denoising Step ($k$): k = 0 (Clean)
Obstacle Radius (m): 0.12 m
Active Trajectory Mode: UPPER_BYPASS Min Clearance: 0.14 m
Safety Status: COLLISION-FREE

3-DOF Articulated Manipulator & Action Chunk Denoising

Click anywhere on workspace to move Green Target Goal • Drag Red Obstacle dynamically

Horizon H=16
Robot Arm (3-DOF IK) Active Action Chunk Parallel DDIM Ensemble (20) Workspace Obstacle
IK Residual: 0.0001 m

Why Diffusion Policy Wins

When human teleoperators avoid an obstacle by going either above or below, a standard MSE neural policy averages both demonstrations—steering the end-effector straight into the obstacle!

DDIM Reverse Denoising:
$$\mathbf{a}^{k-1} = \sqrt{\bar{\alpha}_{k-1}}\,\hat{\mathbf{a}}_0 + \sqrt{1 - \bar{\alpha}_{k-1}}\,\boldsymbol{\epsilon}_\theta(\mathbf{a}^k, \mathbf{o}_t, k)$$
Damped Least Squares IK:
$$\Delta\mathbf{q} = \mathbf{J}^T(\mathbf{J}\mathbf{J}^T + \lambda^2 \mathbf{I})^{-1}\Delta\mathbf{x}$$
Live Joint State (Radians)
Base Shoulder ($\theta_1$):0.42 rad
Elbow Joint ($\theta_2$):-0.85 rad
Wrist Pitch ($\theta_3$):0.31 rad
End-Effector $(x, y)$:(0.20, 0.00)