Instructions to run sample scene?

[vivekgr92]

@jonathanembleyriches Hi Do you have the MCJF and the instructions to run a demo as shown in this video, I want to replicate it on linux

http://youtube.com/watch?v=J_suVHiLBX4

[jonathanembleyriches]

Hi,

The robot MJCFs are sourced from the MuJoCo menagerie (link) and the RoboJuDo repos. The surrounding scenes are made from "raw" UE assets from FAB. Due to the scenes containing paid content I don't think I am allowed to provide the "raw" UE scenes with this FAB content. But, you should be able to reproduce similar ones with the MJCFs from the MuJoCo menagerie and free content from FAB. For the full release I hope to have a sample project containing levels that I can share.

w.r.t the policies, our companion python bridge (urlab_bridge) can run the locomotion and dancing controllers. Our docs give some basic instructions for that here. However, I have a set of PRs queued that overhaul the communication layer in URLab and our Python companion package urlab_bridge. The goal is that anything you'd normally do in the editor (scene assembly, XML import, PIE control, runtime control, and general scene editing) can be driven from Python instead. This includes remote stepping with three modes (live, direct, and puppet). There's also a gym-style API on top. Communication is overhauled to a faster, namespaced wire protocol that the Python side can introspect, so the bridge knows the full surface before any call goes out. These PRs make using the plugin easier and also make deploying policies faster (and more robust).

I'm currently finishing internal testing before the PRs are public, hoping to land them in the next few days. I would suggest waiting for these PRs to land before diving too deep into using the python bridge.

[vivekgr92]

Thanks @jonathanembleyriches Let me know when you push the PR. I will then try to run that on Linux