OmniGibson: a platform for accelerating Embodied AI research built upon NVIDIA's Omniverse engine. Join our Discord for support: https://discord.gg/bccR5vGFEx
The problem here is our get_shortest_path function returns None when no path is found but this is directly fed into a numerical computation for rewards: (self._potential - new_potential)*self._r_potential.
The proposed fix here is to pass in a float no_path_reward arg. I'm not super familiar with this RL infra though, would appreciate input on where this no_path_reward should be exposed to the users.
Original issue: https://github.com/StanfordVL/OmniGibson/issues/715
The problem here is our
get_shortest_pathfunction returnsNonewhen no path is found but this is directly fed into a numerical computation for rewards:(self._potential - new_potential)*self._r_potential.The proposed fix here is to pass in a float
no_path_rewardarg. I'm not super familiar with this RL infra though, would appreciate input on where thisno_path_rewardshould be exposed to the users.