Challenges of Context and Time in Reinforcement Learning: Introducing\n Space Fortress as a Benchmark

Research in deep reinforcement learning (RL) has coalesced around improving\nperformance on benchmarks like the Arcade Learning Environment. However, these\nbenchmarks conspicuously miss important characteristics like abrupt\ncontext-dependent shifts in strategy and temporal sensitivity that are often\npresent in real-world domains. As a result, RL research has not focused on\nthese challenges, resulting in algorithms which do not understand critical\nchanges in context, and have little notion of real world time. To tackle this\nissue, this paper introduces the game of Space Fortress as a RL benchmark which\nincorporates these characteristics. We show that existing state-of-the-art RL\nalgorithms are unable to learn to play the Space Fortress game. We then confirm\nthat this poor performance is due to the RL algorithms' context insensitivity\nand reward sparsity. We also identify independent axes along which to vary\ncontext and temporal sensitivity, allowing Space Fortress to be used as a\ntestbed for understanding both characteristics in combination and also in\nisolation. We release Space Fortress as an open-source Gym environment.\n