A Distributional View on Multi-Objective Policy Optimization
This work addresses a practical problem for practitioners in robotics and reinforcement learning by providing a scale-invariant method for setting preferences, though it appears incremental as it builds on existing multi-objective optimization concepts.
The paper tackles the challenge of expressing numerical preferences for multiple objectives with different units in multi-objective reinforcement learning by proposing a novel algorithm that learns action distributions for each objective and combines them with supervised learning, demonstrating effectiveness on high-dimensional robotics tasks and enabling tracing of nondominated solutions.
Many real-world problems require trading off multiple competing objectives. However, these objectives are often in different units and/or scales, which can make it challenging for practitioners to express numerical preferences over objectives in their native units. In this paper we propose a novel algorithm for multi-objective reinforcement learning that enables setting desired preferences for objectives in a scale-invariant way. We propose to learn an action distribution for each objective, and we use supervised learning to fit a parametric policy to a combination of these distributions. We demonstrate the effectiveness of our approach on challenging high-dimensional real and simulated robotics tasks, and show that setting different preferences in our framework allows us to trace out the space of nondominated solutions.