Creating Autonomous Multi-Object Safe Control via Different Forms of Neural Constraints of Dynamic Programming

Abstract

The aim of this work, which is an extension of previous research, is a comparative analysis of the results of the dynamic optimization of safe multi-object control, with different representations of the constraints of process state variables. These constraints are generated with an artificial neural network and take movable shapes in the form of a parabola, ellipse, hexagon, and circle. The developed algorithm allows one to determine a safe and optimal trajectory of an object when passing other multi-objects. The obtained results of the simulation tests of the algorithm allow for the selection of the best representation of the motion of passing objects in the form of neural constraints. Moreover, the obtained characteristics of the sensitivity of the object’s trajectory to the inaccuracy of the input data make it possible to select the best representation of the motion of other objects in the form of an excessive approximation area as neural constraints of the control process.

Maximum Principle in Autonomous Multi-Object Safe Trajectory Optimization

“…The task of optimizing the control process described above can be solved using many dynamic optimization methods. To solve the problem of optimal multi-object control, the Bellman dynamic programming method described in [18] and Pontryagin's maximum principle are most suitable.…”