RL training time?

[jloganolson]

This repo is awesome! Thank you so much for sharing it. I got training running per the readme.md in HomieRL and it looks like the ETA for completion is just over 100 hours (~4 days) for 100k iterations on 1x4090 - does that sound right?

I couldn't find anything about training time in the paper or repo so thought I'd ask given that's quite a bit higher than, for instance, the 3 hours of PIM (https://arxiv.org/abs/2411.14386) - maybe less iterations/steps?

Thanks again for releasing this!

[Elgce]

Hi, the overall time is for the max_iteration; however, 6k / 8k iters is enough to train a workable policy, so actually training time is much lower.