We model episode sessions--parts of the episode where the latent state isfixed--and propose three keymodifications toexisting meta-RL methods: (i) consistency of latent information within sessions, (ii) session masking, and (iii) priorlatent conditioning.

We study off-policy evaluation (OPE) in partially observable environments with complex observations, with the goal of developing estimators whose guarantee avoids exponential dependence on the horizon.

Autonomous agents operating in real-world scenarios frequently encounter uncertainty and make decisions based on incomplete information.

We also demonstrate its effectiveness in generating robotic tasks.

C Results in ALFRED's T ask and New T asks

Partial observability of the underlying states generally presents significant challenges for reinforcement learning (RL).