A JAX benchmark suite of nine memory-improvable partially observable RL environments, with evidence that recurrent and transformer agents beat memoryless baselines but fall short of state-augmented ceilings.
At the beginning of every episode, both agent start state and goal state are randomly initialized
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Benchmarking Partial Observability in Reinforcement Learning with a Suite of Memory-Improvable Domains
A JAX benchmark suite of nine memory-improvable partially observable RL environments, with evidence that recurrent and transformer agents beat memoryless baselines but fall short of state-augmented ceilings.