Residual Q-Learning: Offline and Online Policy Customization without Value

Buobe IA context · why it matters

Offline and online customization capability — Enables policies learned from demonstrations to adapt to various tasks while preserving their original behavior.
Formulation of policy customization problem — Provides a structured approach for balancing imitation learning with downstream task requirements, facilitating versatile application scenarios. Fique de olho
Customizability in imitation learning will expand its applicability across diverse real-world tasks.

Published by

on •1 min read

Published in: Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS), 2023