Motivated developmental learning of mobile robots in dynamic collision-avoidance

In dynamic collision-avoidance of the environmental cognition, except the expected uncertainty, mobile robots may also encounter unexpected uncertainty. Studying how to deal with the unexpected uncertainty efficiently and flexibly is an important challenge for mobile robots. At present, there are relatively few studies on this aspect, and the mobile robots based on these studies generally lack the ability of autonomous learning, and it is difficult to quickly and flexibly respond to the abrupt external environment. In this paper, a novel collision risk index is designed, which not only considers the influence of the distance of the obstacle, but also that of the speed of the obstacle on the motion of the mobile robots. Simulating the reaction mechanism of the acetylcholine and norepinephrine in human brain in response to environmental uncertainty, through the collision risk index, the attention network of the mobile robot is guided to switch between the dorsal attention network which focuses on the expected stimulus, and the ventral attention network which focuses on new stimulus, and make the robot flexible response to the uncertainty in the environment. At the same time, a new neuronal learning rate is designed to enhance the learning ability of neurons in the hidden layer of the motivated developmental network and improve the robot’s ability to respond quickly to the abrupt environment. In addition, the synaptic weight updating rule is modified to improve the accuracy of the mobile robot’s behavioral decision. Simulation results in two different scenarios, as well as the physical experiment, verify the feasibility of the proposed motivated development learning method in response to unexpected uncertainty in the environment. © 2023 Northeast University. All rights reserved.