Autonomous vehicles learn safer parking in complex urban spaces
Learning Reliable Parking Policies via Offline Reinforcement Learning with Quantized Action Representations
Robotics
Summary
Parking is a tricky but important task for self-driving cars, especially in crowded and messy city parking lots. The authors created a way for these cars to learn reliable parking moves by studying many expert examples and using a clever method to simplify the possible actions into small parts. Their approach helps the car plan and park safely even around other cars that might move unpredictably. Tests in a detailed driving simulator showed this new method worked better than others and handled new parking spots well.
What this means in practice
- •For autonomous vehicle engineers: Develop safer parking algorithms that can handle complex, crowded urban parking environments with uncertain vehicle interactions.
- •For simulation software developers: Integrate interaction-aware parking scenarios and expert data augmentation techniques into vehicle simulators for realistic training.
Authors
Zewei Yang, Zengqi Peng, Jun Ma
Abstract
Parking is a routine yet safety-critical task for autonomous vehicles operating in urban environments. However, cluttered and weakly structured parking spaces, compounded by the interactive uncertainty from surrounding vehicles, hinder reliable maneuver generation. To address these challenges, we develop a waypoint-level offline reinforcement learning framework for interaction-aware autonomous parking. Specifically, a dedicated parking dataset is constructed from hierarchical expert rollouts with rotational waypoint augmentation, covering both non-interactive scenarios and interactive ones. The policy is then conditioned on a compact state representation, in which LiDAR-based obstacle features are adapted to the target pose via feature-wise linear modulation. A state-conditioned tokenizer further quantizes continuous waypoint sequences into discrete action tokens, over which conservative Q-learning is performed to suppress value overestimation on poorly supported actions. Extensive closed-loop experiments are conducted in the high-fidelity CARLA simulator. The proposed framework attains the highest parking success rate among all baselines and transfers reliably to unseen parking slots.