Integral Reinforcement Learning

Integral Reinforcement Learning (IntRL) is a continuous-time control method aiming to efficiently solve optimal control problems by integrating the utility function over time. Current research focuses on improving computational efficiency through advanced quadrature rules and addressing challenges like numerical conditioning and dimensional scaling, often employing model-free approaches and neural network approximations. IntRL's ability to handle uncertain, nonlinear systems with guaranteed stability makes it particularly promising for real-world applications such as robotic control and traffic management, offering robust and data-efficient solutions. The development of decentralized algorithms and improved convergence guarantees are key areas of ongoing investigation.

Papers

February 27, 2024

Impact of Computation in Integral Reinforcement Learning for Continuous-Time Control
Wenhan Cao, Wei Pan
Global Impact Computation Method Continuous Control Bayesian Quadrature Quadrature Rule Quadrature Method Integral Reinforcement Learning

July 18, 2023

Continuous-Time Reinforcement Learning: New Design Algorithms with Theoretical Insights and Performance Guarantees
Brent A. Wallace, Jennie Si
Reinforcement Learning Optimal Control Theoretical Analysis Nonlinear Control Performance Guarantee Integral Reinforcement Learning

March 17, 2023

A Data-Driven Model-Reference Adaptive Control Approach Based on Reinforcement Learning
Mohammed Abouheaf, Wail Gueaieb, Davide Spinello, Salah Al-Sharhan
Reinforcement Learning Data Driven Adaptive Control Unknown Dynamic Model Reference Adaptive Integral Reinforcement Learning

March 15, 2023

Real-Time Measurement-Driven Reinforcement Learning Control Approach for Uncertain Nonlinear Systems
Mohammed Abouheaf, Derek Boase, Wail Gueaieb, Davide Spinello, Salah Al-Sharhan
Reinforcement Learning Nonlinear Dynamic Trajectory Optimization Problem Uncertain Dynamic Actuation Strategy Integral Reinforcement Learning

November 23, 2022

Integral Continual Learning Along the Tangent Vector Field of Tasks
Tian Yu Liu, Aditya Golatkar, Stefano Soatto, Alessandro Achille
Catastrophic Forgetting New Task Vector Field Replay Based Continual Learning Class Incremental Lightweight Continual Integral Reinforcement Learning

September 13, 2022

Data efficient reinforcement learning and adaptive optimal perimeter control of network traffic dynamics
C. Chen, Y. P. Huang, W. H. K. Lam, T. L. Pan, S. C. Hsu, A. Sumalee, R. X. Zhong
Raw Data Efficient Reinforcement Learning Network Traffic Traffic Dynamic Integral Reinforcement Learning Perimeter Defense Problem

Integral Reinforcement Learning

Papers

Impact of Computation in Integral Reinforcement Learning for Continuous-Time Control

Continuous-Time Reinforcement Learning: New Design Algorithms with Theoretical Insights and Performance Guarantees

A Data-Driven Model-Reference Adaptive Control Approach Based on Reinforcement Learning

Real-Time Measurement-Driven Reinforcement Learning Control Approach for Uncertain Nonlinear Systems

Integral Continual Learning Along the Tangent Vector Field of Tasks

Data efficient reinforcement learning and adaptive optimal perimeter control of network traffic dynamics