Muted demo loop from the projectCredit: Diffusion Policy repo (Chi et al.)

Train a Diffusion Policy on your own robot demos

Sim, Medium

Diffusion Policy models robot actions with a denoising diffusion process and is one of the strongest imitation-learning baselines. The repo has Colab notebooks to try it in minutes.

View build on GitHub ↗Share on X

Start with the state or vision Colab, then train on sim tasks like PushT. LeRobot also includes a Diffusion Policy implementation for SO-101 datasets.

Hardware & cost: A GPU (Colab works for the demos).

Difficulty: Medium.

Agent install prompt (paste into Claude Code, Cursor or Codex): "Open the Diffusion Policy Colab linked in https://github.com/real-stanford/diffusion_policy, then reproduce PushT training locally with LeRobot's diffusion policy."

Source: real-stanford/diffusion_policy on GitHub — https://github.com/real-stanford/diffusion_policy (facts checked against the project README, Oct 2026; prices vary by region).

Quick answers

What do you need to train a Diffusion Policy on your own robot demos?: A GPU (Colab works for the demos).
How hard is it to train a Diffusion Policy on your own robot demos?: Medium.

How your agent drives it

Copied from the real-stanford/diffusion_policy README

Python SDK

Shell

$ conda env create -f conda_environment.yaml

Shell

$ python train.py --config-dir=. --config-name=image_pusht_diffusion_policy_cnn.yaml training.seed=42 training.device=cuda:0 hydra.run.dir='data/outputs/${now:%Y.%m.%d}/${now:%H.%M.%S}_${name}_${task_name}'

Any coding agent that can run a shell can run these. Read the README before you run anything on a robot that moves.

Ideas to try with it

Our ideas, not the builder's claims.

Train on the Push-T task in Colab before you own a robot
Swap in your own SO-101 dataset and compare it to ACT
Have your agent launch training and post the eval score when it finishes

1 more idea

Use it as a baseline for a class project

Source: real-stanford/diffusion_policy on GitHub
Last commit: 2024-12-24
Builder: real-stanford
Repo: real-stanford/diffusion_policy, 4,701 stars
Built with: Simulation, SO-100/SO-101 arm (LeRobot)

What to build next

Builds that share a board, robot or agent with this one.