Learned Environment Model

Learned environment models are computational representations of an agent's surroundings, built from data, to improve the efficiency and robustness of reinforcement learning and related tasks. Current research focuses on using these models for tasks like preference elicitation in offline reinforcement learning, continual learning in dynamic environments, and improving the efficiency of gait synthesis in robotics. These models are proving valuable in addressing challenges such as partial observability, limited data, and the need for efficient exploration in complex scenarios, ultimately advancing the capabilities of autonomous agents in various applications.

Papers

June 26, 2024

Preference Elicitation for Offline Reinforcement Learning
Alizée Pace, Bernhard Schölkopf, Gunnar Rätsch, Giorgia Ramponi
Reinforcement Learning Offline Reinforcement Learning Preference Feedback Preference Based Reinforcement Learning Preference Elicitation Learned Environment Model Offline Preference Based Reinforcement Learning

March 12, 2024

Online Continual Learning For Interactive Instruction Following Agents
Byeonghwi Kim, Minhyuk Seo, Jonghyun Choi
Continual Learning Agent Smith Online Continual Learning Embodied Agent Agent Learning Interactive Instruction Learned Environment Model

February 6, 2024

Environment-Centric Learning Approach for Gait Synthesis in Terrestrial Soft Robots
Caitlin Freeman, Arun Niddish Mahendran, Vishesh Vikas
Optimal Gait Dynamic Gait Soft Legged Robot Learned Environment Model

October 25, 2023

Model predictive control-based value estimation for efficient reinforcement learning
Qizhen Wu, Kexin Liu, Lei Chen
Reinforcement Learning Full Model Predictive Control Obstacle Avoidance Efficient Reinforcement Learning Learned Environment Model

July 26, 2022

Repeated Environment Inference for Invariant Learning
Aayush Mishra, Anqi Liu
Invariant Learning Invariant Risk Minimization Learned Environment Model

June 23, 2022

Reinforcement Learning under Partial Observability Guided by Learned Environment Models
Edi Muskardin, Martin Tappler, Bernhard K. Aichernig, Ingo Pill
Reinforcement Learning Markov Decision Process Q Learning Observable Markov Decision Process Partial Observability Learned Environment Model

December 29, 2021

Fully Distributed Informative Planning for Environmental Learning with Multi-Robot Systems
Dohyun Jang, Jaehyun Yoo, Clark Youngdong Son, H. Jin Kim
Reinforcement Learning Task Planning Multi Robot System Exploration Policy Learned Environment Model

Learned Environment Model

Papers

Preference Elicitation for Offline Reinforcement Learning

Online Continual Learning For Interactive Instruction Following Agents

Environment-Centric Learning Approach for Gait Synthesis in Terrestrial Soft Robots

Model predictive control-based value estimation for efficient reinforcement learning

Repeated Environment Inference for Invariant Learning

Reinforcement Learning under Partial Observability Guided by Learned Environment Models

Fully Distributed Informative Planning for Environmental Learning with Multi-Robot Systems