---
datasets:
- commaai/commabody
pipeline_tag: robotics
---

This model has been trained on a larger version (194 minutes total) of the commabody dataset.   
It includes a [vqgan](https://github.com/CompVis/taming-transformers) encoder/decoder fine tuned from imagenet. It compresses images of size 160x256 to 10x16 tokens.  

It also includes a GPT2 model trained to predict the next frame, wheel speeds and actions. It can be used either as a simulator or as a policy. More details in [our blog post](https://blog.comma.ai/a-drive-in-the-office/).

<video title="imagined rollouts from 3s of context" controls>
  <source src="https://blog.comma.ai/img/body_patrol/rollouts.webm" type="video/webm">
</video>

You can run it on a comma body using [our example script in body-jim](https://github.com/commaai/body-jim/blob/master/examples/roam.py).