Suite coverage
We track a 57-game subset spanning dense-reward shooters, sparse puzzles, and multi-objective survival scenarios.
Benchmark
Real-time arcade environments that reward fast reactions, adaptive exploration, and stable policy control.
We track a 57-game subset spanning dense-reward shooters, sparse puzzles, and multi-objective survival scenarios.
Improving long-horizon memory in exploration-heavy games like Montezuma's Revenge while maintaining sample efficiency.