CARL: A Benchmark for Contextual and Adaptive Reinforcement Learning

Abstract

While Reinforcement Learning has made great strides towards solving ever more complicated tasks, many algorithms are still brittle to even slight changes in their environment. This is a limiting factor for real-world applications of RL. Although the research community continuously aims at improving both robustness and generalization of RL algorithms, unfortunately it still lacks an open-source set of well-defined benchmark problems based on a consistent theoretical framework, which allows comparing different approaches in a fair, reliable and reproducible way. To fill this gap, we propose CARL, a collection of well-known RL environments extended to contextual RL problems to study generalization. We show the urgent need of such benchmarks by demonstrating that even simple toy environments become challenging for commonly used approaches if different contextual instances of this task have to be considered. Furthermore, CARL allows us to provide first evidence that disentangling representation learning of the states from the policy learning with the context facilitates better generalization. By providing variations of diverse benchmarks from classic control, physical simulations, games and a real-world application of RNA design, CARL will allow the community to derive many more such insights on a solid empirical foundation.

Reinforcement Learning in Presence of Discrete Markovian Context Evolution

“…Furthermore, we present additional experiments in Appendix suggesting that our model can potentially generalize to non-Markovian and state-dependent settings. While we presented several experiments in various environments, further experimental evaluation is required, e.g., using Benjamins et al (2021).…”

Affordance-Based Human–Robot Interaction With Reinforcement Learning

“…In the literature, several methods exist that include explicit contextual information. For example, Benjamins et al [41], [42] introduced a framework designed to solve CMDPs and a benchmark library. The framework includes information such as gravity, target distance, actuator strength, and joint stiffness in the learning process.…”