Abstract: Learning-based approaches hold great potential for autonomous urban driving motion planning. Compared to traditional rule-based methods, they offer greater flexibility in planning safe and ...