Push-T Diffusion Policy for a Franka Panda with a stick (simulation)
Run dp_image, checkpoint 050000. A LeRobot Diffusion Policy that pushes a T-shaped block onto a fixed
T-shaped target with a stick held by a Franka Panda, in MuJoCo.
- Inputs:
observation.images.overhead,observation.images.wrist,observation.state(images overhead 96×96, wrist 96×96) - Output: commanded stick-tip
[x, y]on the table (metres) at 10 Hz - Data: human mouse-teleoperated demos in simulation
| Sampler | Episodes | Success (≥ 90 % coverage for 0.5 s) |
|---|---|---|
| DDPM (training default) | 50 | 82 % |
pusht_meta.json stores the fps, camera definitions, image resolution and physics XML of the training data,
so the simulator can reproduce the exact training conditions. Code, simulator and usage: see the
project's GitHub repository (scripts/09_download_model.py, demo/sim_demo.py).
- Downloads last month
- 13