Benchmark for Out-of-Distribution Detection in Deep Reinforcement Learning
This work addresses the problem of RL policy failures due to OOD inputs for researchers and practitioners, but it is incremental as it focuses on benchmarking rather than novel detection methods.
The paper tackles the lack of benchmarks for out-of-distribution (OOD) detection in deep reinforcement learning by proposing a new benchmark that modifies physical parameters or corrupts observations in standard environments, and finds that ensemble methods achieve the best OOD detection performance with lower standard deviation across environments.
Reinforcement Learning (RL) based solutions are being adopted in a variety of domains including robotics, health care and industrial automation. Most focus is given to when these solutions work well, but they fail when presented with out of distribution inputs. RL policies share the same faults as most machine learning models. Out of distribution detection for RL is generally not well covered in the literature, and there is a lack of benchmarks for this task. In this work we propose a benchmark to evaluate OOD detection methods in a Reinforcement Learning setting, by modifying the physical parameters of non-visual standard environments or corrupting the state observation for visual environments. We discuss ways to generate custom RL environments that can produce OOD data, and evaluate three uncertainty methods for the OOD detection task. Our results show that ensemble methods have the best OOD detection performance with a lower standard deviation across multiple environments.