Action Noise

Action noise, the addition of random perturbations to actions in reinforcement learning (RL), is a crucial aspect of exploration and efficient policy learning, particularly in continuous control tasks. Current research focuses on optimizing the type, scale, and scheduling of this noise, investigating methods like Gaussian and Ornstein-Uhlenbeck noise, and exploring how to structure noise for improved exploration in complex systems, including the use of latent space manipulation. These investigations aim to enhance RL algorithm performance, particularly in challenging scenarios with high-dimensional action spaces or noisy environments, leading to more robust and efficient learning in robotics, control systems, and other applications. Improved understanding of action noise promises to significantly advance the capabilities of RL agents in real-world settings.

Papers

May 29, 2024

Preferred-Action-Optimized Diffusion Policies for Offline Reinforcement Learning
Tianle Zhang, Jiayi Guan, Lin Zhao, Yihang Li, Dongjiang Li, Zecui Zeng, Lei Sun, Yue Chen, Xuelong Wei, Lusong Li, Xiaodong He
Offline Reinforcement Learning Optimal Policy Offline Reinforcement Learning Algorithm Diffusion Policy Action Noise

May 24, 2024

Model-free reinforcement learning with noisy actions for automated experimental control in optics
Lea Richtmann, Viktoria-S. Schmiesing, Dennis Wilken, Jan Heine, Aaron Tranter, Avishek Anand, Tobias J. Osborne, Michèle Heurs
Reinforcement Learning Partial Observability Optical System Action Noise

December 22, 2023

Noise Morphing for Audio Time Stretching
Eloi Moliner, Leonardo Fierro, Alec Wright, Matti Hämäläinen, Vesa Välimäki
Speech Resynthesis Time Stretching Action Noise

May 31, 2023

Latent Exploration for Reinforcement Learning
Alberto Silvio Chiappa, Alessandro Marin Vargas, Ann Zixiang Huang, Alexander Mathis
Reinforcement Learning Pure Exploration State Perturbation Action Noise Latent Exploration

June 8, 2022

Action Noise in Off-Policy Deep Reinforcement Learning: Impact on Exploration and Performance
Jakob Hollenstein, Sayantan Auddy, Matteo Saveriano, Erwan Renaudo, Justus Piater
Reinforcement Learning Global Impact System Performance Environment Exploration State Space Policy Deep Reinforcement Learning Action Noise

Action Noise

Papers

Preferred-Action-Optimized Diffusion Policies for Offline Reinforcement Learning

Model-free reinforcement learning with noisy actions for automated experimental control in optics

Noise Morphing for Audio Time Stretching

Latent Exploration for Reinforcement Learning

Action Noise in Off-Policy Deep Reinforcement Learning: Impact on Exploration and Performance