A Neural Network Approach to Real-Time Trajectory Generation *

摘要: A neural network approach is proposed for real-time trajectory generation with collision free in an environment varying obstacles and moving target. This biologically inspired topologically organised. The dynamics of each neuron characterised by a shunting equation or additive equation. Each has only local connections, the optimal trajectories are generated without any explicitly optimising cost functions learning. Therefore model computationally eficient. stability analytically proved using Lyapunou function candidate. As examples, applied to formation mobile robot solving maze-type problems, dynamically trucking target, avoiding obstacle. eficiency demonstrated through simulation comparison studies.