UCB-driven Utility Function Search for Multi-objective Reinforcement Learning

TL;DR

This paper introduces a UCB-based method for efficiently searching utility functions in multi-objective reinforcement learning, improving Pareto front quality across benchmarks.

Contribution

It proposes a novel UCB-driven approach for utility function search in MORL, enhancing Pareto front approximation during learning.

Findings

Consistent strong performance on Mujoco benchmarks

Effective maximization of Pareto front hypervolume

Outperforms baseline methods in experiments

Abstract

In Multi-objective Reinforcement Learning (MORL) agents are tasked with optimising decision-making behaviours that trade-off between multiple, possibly conflicting, objectives. MORL based on decomposition is a family of solution methods that employ a number of utility functions to decompose the multi-objective problem into individual single-objective problems solved simultaneously in order to approximate a Pareto front of policies. We focus on the case of linear utility functions parametrised by weight vectors w. We introduce a method based on Upper Confidence Bound to efficiently search for the most promising weight vectors during different stages of the learning process, with the aim of maximising the hypervolume of the resulting Pareto front. The proposed method demonstrates consistency and strong performance across various MORL baselines on Mujoco benchmark problems. The code is released in: https://github.com/SYCAMORE-1/ucb-MOPPO