Abstract
In this paper, we propose a functional extension of the previous sufficient condition to design a larger family of sensor-based path-planning algorithms running in an uncertain world. In previous sensor-based path-planning algorithms, an automaton basically goes straight to its goal, and if the behavior is interfered by an unknown obstacle, the automaton avoids the obstacle by the clockwise or counter-clockwise order. Then the automaton goes straight to the goal again after leaving the obstacle from an arbitrary point which comes close to the goal asymptotically based on the Euclidean distance. This is the previous sufficient condition for keeping the deadlock-free characteristic in the sensor-based path-planning. This paper generalizes the sufficient condition by replacing the Euclidean distance with different types of distances toward the goal. By using an arbitrary distance function flexibly, we get a wider set of sensor-based path-planning algorithms. As an example of them, we show a learning sensor-based path-planning algorithm based on adaptive fitting between the distance function and world shape by sequential tests from the start. By the adaptive fitting, the functional learning algorithm (FLA) makes a shorter deadlock-free path in almost all worlds. Finally the goodness of the learning algorithm and the superiority of the new sufficient condition are ascertained in a graphics simulation.
