Motion planning using adaptive random walks