Matthew Vandergrift

Position: RL Researchers Need to Distinguish Between Solving Simulators and Using Simulators as a Proxy

Matthew Vandergrift, Esraa Elelimy, Martha White

One goal in reinforcement learning (RL) research is to understand general-purpose sequential decision-making, using benchmark simulators as a proxy for learning in deployment settings. When running experiments, however, the goal of achieving high performance in the simulator can…Read more