SEG-Parking: Towards Safe, Efficient, and Generalizable Autonomous Parking Via End-To-End Offline Reinforcement Learning
Zewei Yang, Zengqi Peng, Jun Ma
AI summary
Problem
Autonomous parking in unstructured, dynamic environments struggles with narrow corridors, missing lane markings, and complex vehicle interactions, while existing rule-based, optimization, and imitation learning methods lack generalizability or require extensive online data.
Approach
The authors propose SEG-Parking, which trains an end-to-end policy using a specialized dataset of interacting parking scenarios, a goal-conditioned state encoder pretrained via behavior cloning, and Conservative Q-Learning with a regularizer to prevent unsafe out-of-distribution actions.
Key results
- Highest target success rate in high-fidelity CARLA closed-loop simulations
- Robust generalization to out-of-distribution parking scenarios
- Construction of a specialized interaction-aware autonomous parking dataset
- Public release of the dataset and end-to-end source code
Why it matters
Provides a scalable, data-efficient solution for safe autonomous parking in complex urban environments, accelerating the deployment of practical self-driving systems.
Abstract
Autonomous parking is a critical component for achieving safe and efficient urban autonomous driving. How- ever, unstructured environments and dynamic interactions pose significant challenges to autonomous parking tasks. To address this problem, we propose SEG-Parking, a novel end-to-end offline reinforcement learning (RL) framework to achieve interaction-aware autonomous parking. Notably, a specialized parking dataset is constructed for parking scenarios, which in- clude those without interference from the opposite vehicle (OV) and complex ones involving interactions with the OV. Based on this dataset, a goal-conditioned state encoder is pretrained to map the fused perception information into the latent space. Then, an offline RL policy is optimized with a conservative regularizer that penalizes out-of-distribution actions. Extensive closed-loop experiments are conducted in the high-fidelity CARLA simulator. Comparative results demonstrate the su- perior performance of our framework with the highest success rate and robust generalization to out-of-distribution parking scenarios. The related dataset and source code are available at https://github.com/Yeulerzzz/SEG-Parking.