Adaptive sampling-based motion planning with a non-conservatively defensive strategy for autonomous driving