DERAIL: Diagnostic Environments for Reward And Imitation Learning
Abstract
The objective of many real-world tasks is complex and difficult to procedurally specify. This makes it necessary to use reward or imitation learning algorithms to infer a reward or policy directly from human data. Existing benchmarks for these algorithms focus on realism, testing in complex environments. Unfortunately, these benchmarks are slow, unreliable and cannot isolate failures. As a complementary approach, we develop a suite of simple diagnostic tasks that test individual facets of algorithm performance in isolation. We evaluate a range of common reward and imitation learning algorithms on our tasks. Our results confirm that algorithm performance is highly sensitive to implementation details. Moreover, in a case-study into a popular preference-based reward learning implementation, we illustrate how the suite can pinpoint design flaws and rapidly evaluate candidate solutions. The environments are available at https://github.com/HumanCompatibleAI/seals .
Cite
@article{arxiv.2012.01365,
title = {DERAIL: Diagnostic Environments for Reward And Imitation Learning},
author = {Pedro Freire and Adam Gleave and Sam Toyer and Stuart Russell},
journal= {arXiv preprint arXiv:2012.01365},
year = {2020}
}