comma world model 4B
- AI model
- Cars
A toolkit for distilling the openpilot driving model into a smaller driving model, trained on logged drive segments and then refined in a learned world model used as a simulator.
Made by comma.ai.
Source on GitHub: github.com/YassineYousfi/openpilot.distill (opens in a new tab). Facts read at commit 294d57a (opens in a new tab).
Licenses
- Code
- MIT (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 294d57a (opens in a new tab), on 16 Oct 2026.
- Commercial use
- Yes
What it does
- Runs a localization pipeline over logged drive segments to produce training targets.
- Trains a smaller driving model using localized targets and distillation targets from the openpilot model.
- Fine-tunes the driving model in a learned world model with DAgger-style post-training.
- Accepts segment video files and rlogs placed in a data folder.
- Can use public videos and labels from the comma1M dataset on Hugging Face if no own logs are available.
Written by Robotheca from the project's README (opens in a new tab) at commit 294d57a.
Getting started
- Difficulty
- Advanced
- How to start
- The project's README (opens in a new tab)
Maintenance
- Status
- Active: a commit in the last 90 days
- Newest commit
- 1 Sept 2026 on master
- Latest release
- None published on GitHub
Checked on GitHub on 16 Oct 2026.
Links
Robots that can run it
No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.
Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.