Open and completed research requests

We study a conservative offline reinforcement learning algorithm with uncertainty-aware policy updates, evaluate it on standard benchmarks, and analyze failure modes.