Loading [a11y]/accessibility-menu.js
Policy Iteration Solution for Differential Games with Constrained Control Policies | IEEE Conference Publication | IEEE Xplore

Policy Iteration Solution for Differential Games with Constrained Control Policies


Abstract:

Graphical games are special classes of the standard differential games. The underlying neural network solutions are complicated and do not employ straightforward tuning l...Show More

Abstract:

Graphical games are special classes of the standard differential games. The underlying neural network solutions are complicated and do not employ straightforward tuning laws. This issue becomes more challenging if the control strategies of the agents are constrained. An integral adaptive learning approach is developed to find an online solution for the differential graphical games with constrained control strategies. This solution employs a distributed adaptive policy iteration process in real-time. Local performance indices are utilized to assess the coupling between the agents and account for the constrained policies. Means of adaptive critics are used to develop a solution platform for each agent using single layer of neural networks., that are adapted using gradient descent tuning approach. This framework handles the main concerns related to the complexity and scalability of the distributed solution. The convergence of the adaptive learning solution is shown to hold under some graph-based conditions.
Date of Conference: 10-12 July 2019
Date Added to IEEE Xplore: 29 August 2019
ISBN Information:

ISSN Information:

Conference Location: Philadelphia, PA, USA

I. Introduction

The constrained graphical games are subcategories [1]–[3] of the standard games, where the interactions between agents on graphs are accounted for to develop online solutions for the dynamic games [2], [3]. The constrained nature of the control signals makes the problem challenging, since the solutions of the constrained graphical games depend on complicated neural network structures and tuning approaches, even if the control inputs are not constrained.

Contact IEEE to Subscribe

References

References is not available for this document.