Goal Exploration

Goal exploration in reinforcement learning focuses on enabling agents to efficiently discover and achieve goals, especially in complex environments with sparse rewards and long time horizons. Current research emphasizes improving exploration strategies, such as leveraging learned skills or employing "post-exploration" techniques where the agent revisits previously successful states before venturing into the unknown, to enhance the discovery of novel subgoals. These advancements are crucial for solving challenging real-world problems where explicitly defining all subgoals is impractical, impacting fields like robotics and autonomous systems.

Papers

November 3, 2024

April 19, 2024

Goal Exploration via Adaptive Skill Distribution for Goal-Conditioned Reinforcement Learning
Lisheng Wu, Ke Chen
Goal Conditioned Reinforcement Learning Exploration Efficiency Goal Directed Behavior Adaptive Skill Goal Exploration

December 6, 2022

First Go, then Post-Explore: the Benefits of Post-Exploration in Intrinsic Motivation
Zhao Yang, Thomas M. Moerland, Mike Preuss, Aske Plaat
Reinforcement Learning Environment Exploration Complementary Benefit Intrinsic Motivation First Attempt POST Exploration Goal Exploration

October 28, 2022

Goal Exploration Augmentation via Pre-trained Skills for Sparse-Reward Long-Horizon Goal-Conditioned Reinforcement Learning
Lisheng Wu, Ke Chen
Pre Trained Goal Conditioned Reinforcement Learning Sub Goal Goal Exploration

Goal Exploration

Papers

Learning World Models for Unconstrained Goal Navigation

Exploring the Edges of Latent State Clusters for Goal-Conditioned Reinforcement Learning

Goal Exploration via Adaptive Skill Distribution for Goal-Conditioned Reinforcement Learning

First Go, then Post-Explore: the Benefits of Post-Exploration in Intrinsic Motivation

Goal Exploration Augmentation via Pre-trained Skills for Sparse-Reward Long-Horizon Goal-Conditioned Reinforcement Learning