Cliff Diving: Exploring Reward Surfaces in Reinforcement Learning Environments
This work provides new visual insights into reinforcement learning optimization landscapes, helping researchers understand and address failure modes in RL agents, though it is incremental in extending visualization techniques to more environments.
The paper generated reward surface visualizations for 27 popular Gym reinforcement learning environments, revealing frequent 'cliffs' (sudden drops in expected return) and showing that A2C often falls into low-reward regions while PPO avoids them, confirming intuitions about PPO's performance.
Visualizing optimization landscapes has led to many fundamental insights in numeric optimization, and novel improvements to optimization techniques. However, visualizations of the objective that reinforcement learning optimizes (the "reward surface") have only ever been generated for a small number of narrow contexts. This work presents reward surfaces and related visualizations of 27 of the most widely used reinforcement learning environments in Gym for the first time. We also explore reward surfaces in the policy gradient direction and show for the first time that many popular reinforcement learning environments have frequent "cliffs" (sudden large drops in expected return). We demonstrate that A2C often "dives off" these cliffs into low reward regions of the parameter space while PPO avoids them, confirming a popular intuition for PPO's improved performance over previous methods. We additionally introduce a highly extensible library that allows researchers to easily generate these visualizations in the future. Our findings provide new intuition to explain the successes and failures of modern RL methods, and our visualizations concretely characterize several failure modes of reinforcement learning agents in novel ways.