Jonathan (Buzz) Embley-Riches

7 posts

Jonathan (Buzz) Embley-Riches

Jonathan (Buzz) Embley-Riches

@JonEmbleyRiches

Current PhD student in Robotics and Artificial Intelligence at UCL.

Katılım Haziran 2023
21 Takip Edilen139 Takipçiler
Jonathan (Buzz) Embley-Riches
Jonathan (Buzz) Embley-Riches@JonEmbleyRiches·
@Shi_Soul Not directly; MJWarp/MJLab are best for parallel RL. URLab embeds MuJoCo for high-fidelity data gen & vision eval. An interesting direction is leveraging UE PCG in URLab to generate DR scenes for MJLab to load for training, then zero-shot eval in URLab. Open to exploring!
English
0
0
2
72
Weiji Xie
Weiji Xie@Shi_Soul·
@JonEmbleyRiches Cool! just wondering, is there any plan to support parallel simulation & RL training?
English
1
0
3
173
Jonathan (Buzz) Embley-Riches
Jonathan (Buzz) Embley-Riches@JonEmbleyRiches·
@JulienBlanchon Thanks! Technically yes, though currently untested. You would package the project for Linux, which is already working, and run via NVIDIA Container Toolkit with the -RenderOffscreen flag. I can create a Dockerfile if there's enough interest.
English
1
0
3
382
Jonathan (Buzz) Embley-Riches
Jonathan (Buzz) Embley-Riches@JonEmbleyRiches·
Alpha release — open source under Apache 2.0. If you are working in robotics simulation, I would love to hear from you. Check the roadmap on GitHub for upcoming features and feel free to open an issue or PR.
English
1
0
9
717
Jonathan (Buzz) Embley-Riches
Jonathan (Buzz) Embley-Riches@JonEmbleyRiches·
Features: - Import or author robots natively - Deploy policies live in-engine - Possess robots with real-time input - Turn any mesh into a physics object - Record, replay, and adjust sim speed - Stream over ZMQ or ROS 2 - Scale to dozens of robots
English
1
0
7
932