Hybridizing RRT and variable-length genetic algorithm for smooth path generation

A smooth path generation scheme based on integrating rapidly-exploring random tree (RRT) with island parallel variable-length genetic algorithm with migration is presented for finding G³-continuous η³-spline paths that minimize a quality measure combining path length and curvature. By injecting RRT solutions into an isolated initial subpopulation and occasional migration to allow the competition and mixture of genetic information between the best individuals of subpopulations whenever feasible paths are discovered in other islands, the hybrid approach discovers smoother path more efficiently while maintaining the diversity of evolutionary optimizer and preventing the premature due to the RRT injection. The simulation results in complex maps demonstrated the advantage of RRT-injection in our implementation, and the hybrid approach is quite flexible and adaptive to generate paths with achievable smoother curvature profile at the expense of a little increased runtime and/or path length.