Online Data-Driven Inverse Reinforcement Learning for Deterministic Systems

In this paper, we present an online data-driven inverse reinforcement learning (IRL) method for estimating the cost function of continuous-time linear and nonlinear deterministic systems from state and input measurements. Our approach utilizes Bellman error, obtained from integral reinforcement lear...

Full description

Saved in:
Bibliographic Details
Published in2022 IEEE Symposium Series on Computational Intelligence (SSCI) pp. 884 - 889
Main Authors Asl, Hamed Jabbari, Uchibe, Eiji
Format Conference Proceeding
LanguageEnglish
Published IEEE 04.12.2022
Subjects
Cost function
cost function estimation
Estimation error
Heuristic algorithms
Integral equations
Inverse reinforcement learning
Measurement
Numerical simulation
online data-driven solution
Reinforcement learning
Online AccessGet full text
DOI10.1109/SSCI51031.2022.10022226

Cover

More Information
Summary:In this paper, we present an online data-driven inverse reinforcement learning (IRL) method for estimating the cost function of continuous-time linear and nonlinear deterministic systems from state and input measurements. Our approach utilizes Bellman error, obtained from integral reinforcement learning, with error derived from the closed-form equation of an optimal controller as the performance metric to develop a recursive IRL technique. Our proposed scheme does not require the time derivative of states or the drift dynamics of a system. We describe a Lyapunov-based analysis to show the ultimate boundedness of the estimation errors. Simulation studies demonstrate the effectiveness of the proposed method.
DOI:10.1109/SSCI51031.2022.10022226