DEPLOYDatabase

Brain

Diffusion Policy

Columbia University's visuomotor policy that formulates robot control as denoising diffusion. Achieves +46.9% improvement over prior methods. Open-source, widely adopted as a baseline for imitation learning.

Research model · Maturity: Research · Open source

Machine-readable surfaces

Architecture

Denoising diffusion for visuomotor control, receding horizon, time-series transformer

Key facts

Method and benchmark claim (primary project page)
The Diffusion Policy project page describes representing a robot's visuomotor policy as a conditional denoising diffusion process. It says the method was benchmarked across 12 tasks from 4 robot manipulation benchmarks with an average improvement of 46.9% over prior state-of-the-art. Authors are affiliated with Columbia University, Toyota Research Institute, and MIT. Venue notes include RSS 2023 and IJRR 2024. Treat the 46.9% figure as author-reported (Diffusion Policy)
Technical contributions (primary project page)
The same page lists key technical contributions including receding horizon control, visual conditioning, and the time-series diffusion transformer, and says diffusion policies handle multimodal action distributions and high-dimensional action spaces with strong training stability. Frame advantages as paper-stated (Diffusion Policy)
Visuomotor action diffusion framing
The Columbia Diffusion Policy project page describes Diffusion Policy as a visuomotor policy-learning method that formulates robot control as a denoising diffusion process over action sequences (Diffusion Policy).
Primary paper host
The project hosts the 2023 Diffusion Policy paper PDF describing visuomotor policy learning via action diffusion for robot manipulation (Diffusion Policy paper).
Visuomotor action diffusion
Diffusion Policy represents visuomotor policies as conditional denoising diffusion processes for robot learning (Diffusion Policy).
Robot learning implementation
The Diffusion Policy project publishes research materials and implementation resources for state-based and vision-based robot policy learning (Diffusion Policy paper).
Open implementation
The project maintains an open implementation with image-based policy modules and task configurations in the real-stanford GitHub repository (Diffusion Policy repository).
Real-world manipulation tasks
The Diffusion Policy materials document real-world robot-learning evaluations including the PushT task (Diffusion Policy paper).

Mentioned on (2)

Registry pages whose text names Diffusion Policy.

Companies (1)

Walden Robotics

Robots and products (1)

Walden Robot

Common questions

What is Diffusion Policy?
Columbia University's visuomotor policy that formulates robot control as denoising diffusion. Achieves +46.9% improvement over prior methods. Open-source, widely adopted as a baseline for imitation learning.
Is Diffusion Policy open source?
Yes. Diffusion Policy is recorded as open-source / open-weights on the DEPLOY registry, meaning model weights or source code are publicly available.
What type of AI is Diffusion Policy?
Diffusion Policy is a research model, built on a Denoising diffusion for visuomotor control, receding horizon, time-series transformer architecture on the DEPLOY registry.
What is Diffusion Policy's maturity stage?
Diffusion Policy is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.
Which robots run on Diffusion Policy?
No robot models on the DEPLOY registry are recorded as running Diffusion Policy. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.

Sources (4)

  1. https://diffusion-policy.cs.columbia.edu/
  2. https://diffusion-policy.cs.columbia.edu/diffusion_policy_2023.pdf
  3. https://diffusion-policy.cs.columbia.edu/diffusion_policy_ijrr.pdf
  4. https://github.com/real-stanford/diffusion_policy
Methodology: Verified · 4 sources (no primary) · last reviewed 2026-10-09

Verification posture

Verified

Low confidence

Review state

Stable

Last reviewed 2026-10-09

Maturity + lifecycle

Maturity stage: research

Sources by quality tier

3
unclassified
Unclassified source
1
code-repository
Code repository

The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.

Methodology surface for Diffusion Policy.

Canonical ID 4bfb24fe-bf26-4ef0-b02a-79df81b7fc87