Quantifying Generalization in Reinforcement Learning

We’re releasing a new training environment, CoinRun, that precisely quantifies an agent’s ability to transfer its experience to novel test environments....