Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Voelcker, Claas, Brunnbauer, Axel, Hussing, Marcel, Nauman, Michal, Abbeel, Pieter, Eaton, Eric, Grosu, Radu, Farahmand, Amir-massoud, Gilitschenski, Igor
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2507.11019
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915930609549312
author Voelcker, Claas
Brunnbauer, Axel
Hussing, Marcel
Nauman, Michal
Abbeel, Pieter
Eaton, Eric
Grosu, Radu
Farahmand, Amir-massoud
Gilitschenski, Igor
author_facet Voelcker, Claas
Brunnbauer, Axel
Hussing, Marcel
Nauman, Michal
Abbeel, Pieter
Eaton, Eric
Grosu, Radu
Farahmand, Amir-massoud
Gilitschenski, Igor
contents Score-function based methods for policy learning, such as REINFORCE and PPO, have delivered strong results in game-playing and robotics, yet their high variance often undermines training stability. Using pathwise policy gradients, i.e. computing a derivative by differentiating the objective function, alleviates the variance issues. However, they require an accurate action-conditioned value function, which is notoriously hard to learn without relying on replay buffers for reusing past off-policy data. We present an on-policy algorithm that trains Q-value models purely from on-policy trajectories, unlocking the possibility of using pathwise policy updates in the context of on-policy learning. We show how to combine stochastic policies for exploration with constrained updates for stable training, and evaluate important architectural components that stabilize value function learning. The result, Relative Entropy Pathwise Policy Optimization (REPPO), is an efficient on-policy algorithm that combines the stability of pathwise policy gradients with the simplicity and minimal memory footprint of standard on-policy learning. Compared to state-of-the-art on two standard GPU-parallelized benchmarks, REPPO provides strong empirical performance at superior sample efficiency, wall-clock time, memory footprint, and hyperparameter robustness.
format Preprint
id arxiv_https___arxiv_org_abs_2507_11019
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Relative Entropy Pathwise Policy Optimization
Voelcker, Claas
Brunnbauer, Axel
Hussing, Marcel
Nauman, Michal
Abbeel, Pieter
Eaton, Eric
Grosu, Radu
Farahmand, Amir-massoud
Gilitschenski, Igor
Machine Learning
Score-function based methods for policy learning, such as REINFORCE and PPO, have delivered strong results in game-playing and robotics, yet their high variance often undermines training stability. Using pathwise policy gradients, i.e. computing a derivative by differentiating the objective function, alleviates the variance issues. However, they require an accurate action-conditioned value function, which is notoriously hard to learn without relying on replay buffers for reusing past off-policy data. We present an on-policy algorithm that trains Q-value models purely from on-policy trajectories, unlocking the possibility of using pathwise policy updates in the context of on-policy learning. We show how to combine stochastic policies for exploration with constrained updates for stable training, and evaluate important architectural components that stabilize value function learning. The result, Relative Entropy Pathwise Policy Optimization (REPPO), is an efficient on-policy algorithm that combines the stability of pathwise policy gradients with the simplicity and minimal memory footprint of standard on-policy learning. Compared to state-of-the-art on two standard GPU-parallelized benchmarks, REPPO provides strong empirical performance at superior sample efficiency, wall-clock time, memory footprint, and hyperparameter robustness.
title Relative Entropy Pathwise Policy Optimization
topic Machine Learning
url https://arxiv.org/abs/2507.11019