LIRA

We work on hierarchical reinforcement learning (HRL) methods to create autonomous agents that can efficiently solve complex tasks. We use hierarchies to improve, for example, sample efficiency in complex navigation tasks. We also work on discovering useful subtasks, with an emphasis on highly dynamic and stochastic environments, like those with adversarial agents. Finally, we showed that hierarchies can improve adaptation to new dynamics.

Relevant publications

Intrinsic Reward Policy Optimization for Sparse-Reward Environments

arXiv'26

IRPO uses multiple exploratory policies to update a base policy.

Feature Specialization and Clustering Improves Hierarchical Sub-task Learning

ALA'22

Value maps for object and spatial options.

Evaluating Adaptation Performance of Hierarchical Deep Reinforcement Learning

ICRA'20

A hierarchical decision policy.

This work was funded in part by DARPA FA8750-18C-0137 and ONR N00014-20-1-2249.