search

Farama-Foundation / Metaworld

Collections of robotics environments geared towards benchmarking multi-task and meta reinforcement learning
https://metaworld.farama.org/
MIT License
1.28k stars 274 forks source link

Safely Using Customized Environment Without Goal/End Effector Sites for Visual-Based Task Completion #491

Closed EwertzJN closed 3 months ago

EwertzJN commented 3 months ago

Question: Safely Using Customized Environment Without Goal/End Effector Sites for Visual-Based Task Completion

Description

I have customized the Meta-World environments for several tasks (Button Press, Hammer, Box Close, Drawer Close) by modifying the visibility of the goal and end effector sites to make them invisible. This was done by setting the alpha value of the RGBA parameter for these sites to 0. My main interest is to solve these tasks purely based on visual input from the rendered images.

Customization Details

Here is the commit with the changes I made to the Meta-World codebase: Commit: Make sites invisible

Question:

Are these modifications safe to use for solving the tasks using only visual input?

reginald-mclean commented 3 months ago

I don’t know. I don’t know your use case is, or what you’re trying to accomplish. I really think that’s up to you to judge