RLxUSD: a minimal convention for recording Reinforcement Learning episodes in OpenUSD

Community USD
1

Hello everyone,
This is my first post here, glad to join the discussion.

I’ve been exploring how Reinforcement Learning (RL) environments could export complete episodes as native OpenUSD scenes.

I developed RLxUSD v0.1, a small convention and Qt-based usdview plugin that uses standard USD prims, timeSamples, and customData to record and visualize full RL trajectories.

The goal is to enable inspection-first workflows in RL research and simulation, without needing multiple output formats like CSV logs or MP4 videos.

If anyone in the OpenUSD or simulation community finds this relevant, I’d appreciate feedback on schema design or interoperability aspects.

https://github.com/dorado-daniel/RLxUSD

Whitepaper:

1 Like
2

Nice work @danieldorado ! I don’t know how to use this myself, but it looks like a good idea. I shared it on LinkedIn so that smarter folks than me can check it out. :slightly_smiling_face:

1 Like