This paper shows how to implement curvature-optimal solutions of path-planning and trajectoryservoing problems for autonomous vehicles by using neural nets. The neural net is used to implement the nonlinear optimal state-feedback laws. The trained net can be used in real-time control. The net is trained by simulated data generated at random. No two-point boundary-value problems need to be solved. The method presented can, in general, be applied to a wide range of time-invariant terminal control or servo control problems.
|Journal||Engineering Applications of Artificial Intelligence|
|Publication status||Published - 1994|
|MoE publication type||A1 Journal article-refereed|