Real-time collision-free motion planning of nonholonomic robots using a neural dynamics based approach

A novel neural dynamics based approach to smooth, continuous and collision-free path generation of an autonomous nonholonomic mobile robot is proposed. The robot behavior, such as target acquisition and obstacle avoidance, are completely controlled by two control variables, the heading direction and the forward velocity of the robot. The dynamics of these control variables is characterized by a biologically inspired shunting neural model, whose inputs are from the target and obstacles that are acquired relying on measurable sensors information only. The target input produces an attractive force, while the obstacle inputs form repulsive forces to the mobile robot. Each force votes for a certain value of control variables that have unique values at a certain time. The collision-free path and the velocity control commands of the robot are generated through the dynamics of control variables. The kinematic constraints of mobile robot is respected. A series of simulation results show that the proposed approach can be successfully applied to both static and dynamic environments, as well as multi-robot systems with effective and efficient computation