Outcome-Driven Reinforcement Learning via Variational Inference
This addresses the challenge of reward shaping for practitioners in robotics and AI, offering a more automated approach to policy learning.
The paper tackles the problem of manually designing reward functions in reinforcement learning by framing it as inferring policies that achieve desired outcomes, and demonstrates that their method eliminates the need for hand-crafted rewards across diverse manipulation and locomotion tasks.
While reinforcement learning algorithms provide automated acquisition of optimal policies, practical application of such methods requires a number of design decisions, such as manually designing reward functions that not only define the task, but also provide sufficient shaping to accomplish it. In this paper, we view reinforcement learning as inferring policies that achieve desired outcomes, rather than as a problem of maximizing rewards. To solve this inference problem, we establish a novel variational inference formulation that allows us to derive a well-shaped reward function which can be learned directly from environment interactions. From the corresponding variational objective, we also derive a new probabilistic Bellman backup operator and use it to develop an off-policy algorithm to solve goal-directed tasks. We empirically demonstrate that this method eliminates the need to hand-craft reward functions for a suite of diverse manipulation and locomotion tasks and leads to effective goal-directed behaviors.