A variational approach to constructivist learning for mobile robot navigation

In this paper, we present a constructivist approach for the learning by example problem, where control laws (or behaviors) are learned in order to approximate a training trajectory. The new behaviors are learned by systematically improving upon existing capabilities. Within this context, the learning problem is formulated as an optimal control problem, and variational arguments are used to obtain optimality conditions. Numerical algorithms that utilize the optimality conditions to attain a stationary solution are produced. A small-scale navigation example is discussed in order to highlight the operation of the proposed approach.