Integraing Algorithm with Different Benchmarks

[vaggelisT]

Hello,

I am interested in integrating this RL algorithm with different benchmark environments. Could you please guide me on whether this is feasible? If yes, what specific changes and additions would I need to implement to achieve this integration successfully?

Thank you.

[nicklashansen]

Yes, this is certainly possible and should not be too much work! You would need to add your environment to the tasks directory here: https://github.com/facebookresearch/modem/tree/main/tasks and collect a few demonstrations for use with MoDem. To collect demonstrations you can either use motion planning, a scripted policy, human teleoperation, or simply train an RL policy with e.g. dense rewards / privileged information.