Abstract

A Hamiltonian based adaptive critic structure is proposed for solving missile guidance problems. This structure consists of a supervisor neural network called 'critic' and a controller network called 'action'. Together they are used to solve model-based guidance problems. The advantage of this approach is that: i) the training data for each network is created by the other network, ii) the converged solutions yield near optimal guidance over the entire span of the training range, and iii) they can be used as feedback controllers though trained off-line. A main contribution is that we establish a direct link between traditional optimal control formulation and neural networks. The resultant network can act as a repository of gains and its structure is very general. Numerical simulations through an illustrative scalar problem and a typical target-intercept problem demonstrate the potential of the approach.

Department(s)

Mechanical and Aerospace Engineering

Publication Status

Open Access

Document Type

Article - Conference proceedings

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2024 American Institute of Aeronautics and Astronautics, All rights reserved.

Publication Date

01 Jan 1996

Share

 
COinS