A Computational Developmental Model of Perceptual Learning for Mobile Robot

During the perceptual learning (PL) of a mobile robot, how to improve the efficiency of its behavior decision is a great challenge. The existing methods suffer from the less flexibility and low efficiency. This article proposes a novel methodology to address these problems. It mimics the principle that the human’s brain can still think, even when he/she neither receives any sensory stimulus nor performs any action, during the off-task process, a gated self-organization mechanism is used to trigger the “brain” of the mobile robot to work, analyze the scenarios that it encountered in the former tasks to decide the best moving direction, then store the sensed scenario information by means of the lateral excitation of the internal neurons in a developmental network (DN), and set up the weight connections from the internal area to high-level decision-making area of the DN. In the following PL, if the robot encounters a similar scenario, it can make a better and faster behavior decision. Therefore, this methodology can greatly enhance the efficiency of behavior decision with the limited training samples, without retraining in facing a new scenario. Most importantly, this methodology makes the robot continuously improve its intelligence through the autonomous learning, even in offline state. Extensive simulation and experiment results of the mobile robot PL demonstrate its potential. Since it is task nonspecific, the same learning principles are potentially suitable for other fields. To the best of our knowledge, it is the first trial to apply the transfer learning through the lateral excitation mechanism of internal neurons to the robot field with the emergent representation.