Inverse Reinforcement Learning with Multiple Planning Horizons

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yao, Jiayu, Pan, Weiwei, Doshi-Velez, Finale, Engelhardt, Barbara E
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913520320249856
author Yao, Jiayu
Pan, Weiwei
Doshi-Velez, Finale
Engelhardt, Barbara E
author_facet Yao, Jiayu
Pan, Weiwei
Doshi-Velez, Finale
Engelhardt, Barbara E
contents In this work, we study an inverse reinforcement learning (IRL) problem where the experts are planning under a shared reward function but with different, unknown planning horizons. Without the knowledge of discount factors, the reward function has a larger feasible solution set, which makes it harder for existing IRL approaches to identify a reward function. To overcome this challenge, we develop algorithms that can learn a global multi-agent reward function with agent-specific discount factors that reconstruct the expert policies. We characterize the feasible solution space of the reward function and discount factors for both algorithms and demonstrate the generalizability of the learned reward function across multiple domains.
format Preprint
id arxiv_https___arxiv_org_abs_2409_18051
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Inverse Reinforcement Learning with Multiple Planning Horizons
Yao, Jiayu
Pan, Weiwei
Doshi-Velez, Finale
Engelhardt, Barbara E
Machine Learning
In this work, we study an inverse reinforcement learning (IRL) problem where the experts are planning under a shared reward function but with different, unknown planning horizons. Without the knowledge of discount factors, the reward function has a larger feasible solution set, which makes it harder for existing IRL approaches to identify a reward function. To overcome this challenge, we develop algorithms that can learn a global multi-agent reward function with agent-specific discount factors that reconstruct the expert policies. We characterize the feasible solution space of the reward function and discount factors for both algorithms and demonstrate the generalizability of the learned reward function across multiple domains.
title Inverse Reinforcement Learning with Multiple Planning Horizons
topic Machine Learning
url https://arxiv.org/abs/2409.18051