Learning What To Do by Simulating the Past

Abstract

Since reward functions are hard to specify, recent work has focused on learning policies from human feedback. However, such approaches are impeded by the expense of acquiring such feedback. Recent work proposed that agents have access to a source of information that is effectively free: in any environment that humans have acted in, the state will already be optimized for human preferences, and thus an agent can extract information about what humans want from the state. Such learning is possible in principle, but requires simulating all possible past trajectories that could have led to the observed state. This is feasible in gridworlds, but how do we scale it to complex tasks? In this work, we show that by combining a learned feature encoder with learned inverse models, we can enable agents to simulate human actions backwards in time to infer what they must have done. The resulting algorithm is able to reproduce a specific skill in MuJoCo environments given a single state sampled from the optimal policy for that skill.

arXiv commentPresented at ICLR 2021

References in corpus (12)

• Auto-Encoding Variational Bayes
• Neural Architecture Search with Reinforcement Learning
• Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
• Neural Architecture Search: A Survey
• Reinforcement Learning and Control as Probabilistic Inference: Tutorial and Review
• Learning Latent Dynamics for Planning from Pixels
• Generative Adversarial Imitation from Observation
• Learning Plannable Representations with Causal InfoGAN
• Learning to Understand Goal Specifications by Modelling Reward
• Learning and Querying Fast Generative Models for Reinforcement Learning
• Learning to summarize from human feedback
• Preferences Implicit in the State of the World