Organizations: Jishu Miao, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi are with Chubu University, Kasugai, Japan. · Han Chen is with Tongji University, Shanghai, China. · Jiankun Zhai, Qi Liu are independent researchers.
Autonomous parking fundamentally differs from on-road driving due to its frequent direction changes and complex maneuvering requirements. However, existing End-to-End (E2E) planning methods often simplify the parking task into a geometric path regression problem, neglecting explicit modeling of the vehicle's kinematic state. This "dimensionality deficiency" easily leads to physically infeasible trajectories and deviates from real human driving behavior, particularly at critical gear-shift points in multi-shot parking scenarios. In this paper, we propose SunnyParking, a novel dual-branch E2E architecture that achieves motion state awareness by jointly predicting spatial trajectories and discrete motion state sequences (e.g., forward/reverse). Additionally, we introduce a Fourier feature-based representation of target parking slots to overcome the resolution limitations of traditional bird's-eye view (BEV) approaches, enabling high-precision target interactions. Experimental results demonstrate that our framework generates more robust and human-like trajectories in complex multi-shot parking scenarios, while significantly improving gear-shift point localization accuracy compared to state-of-the-art methods. We open-source a new parking dataset of the CARLA simulator, specifically designed to evaluate full prediction capabilities under complex maneuvers.
Figures & tables
Fig. 1 : Comparison of parking planning paradigms. Geometric Plan represents many current E2E approaches by focusing primarily on geometric features. Modular Plan represents the traditional rule-based paradigm (e.g., Hybrid A*). While it generates a kinematic feasible plan, achieving human-like planning remains a challenge. Our Unified Kinematic Plan achieves comparable feasibility through explicit kinematic modeling, while demonstrating superior “Human-like” qualities via its learning-based approach.
Fig. 2 : The Overiew of SunnyParking’s Network Architecture . It takes 4 RGB images as the sensory input, along side it is the selected target slot. Vision Encoder extracts the vision-domain features then project them to the BEV space as the spatial features. Fourier Mapping Encoder first represents the target including it’s (x,y,θ) in Fourier feature, then query the spatial features. In YUKUI module: Trajectory Branch takes the fused features then predicts a set of waypoints. With the additional input from Trajectory Branch’s hidden layer, Motion State Branch predicts the motion state, corresponding with each points
Fig. 3 : The Architecture of Motion State Branch. Motion Query is enriched by the trajectory feature via cross-attention. This resulting the fused Queries is then input to a Transformer Decoder.
Fig. 4 : Progressive transition from teacher forcing to autoregressive prediction via Scheduled Sampling.
TABLE I: Performance Evaluation of Trajectory and Gear Shift Points.
Set
Input Rep.
Model Comp.
Sched. Sampling
Trajectory Metrics
Motion State Metrics
BEV
Fourier
Traj.
Motion
L2 Dis.
Four. Diff.
Haus. Dis.
AHE
Acc.
Avg. Dist.
1
✓
✓
3.505
40.033
3.760
/
/
/
2
✓
✓
✓
2.327
39.492
2.408
/
/
/
3
✓
✓ext
✓
1.474
27.253
1.635
9.8
0.827
1.900
4
✓
✓ext
✓
✓
1.800
37.253
2.160
11.7
0.794
2.385
5
✓
✓
2.863
55.885
3.172
/
/
/
TABLE II : Ablation Study of Different Model Components.
Fig. 5 : Qualitative Comparison of Parking Planning Results. Blue represents the GT trajectory, while red indicates the prediction. As task complexity increases, baseline methods struggle: TransFuser fails to generate feasible trajectories, while ParkingE2E exhibits overfitting to specific parking patterns. Our method produces robust trajectories with correctly predicted gear shift points. Notably, as shown in ( 5(o) ), our method demonstrates superior feature representation capabilities. It effectively handles regions near the boundaries of the BEV FOV, comparable to the direct 2D coordinate input strategy used in TransFuser.
Fig. 6 : Comparison of Attention Heatmaps between Baseline (BEV) and Ours (Fourier). Our method exhibits an attention shift as the relative position between the ego rear and the target slot changes.
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Fig. 7 : Data Collection Scenario of the Parking Lot in BEV. The orange boxes represent the 16 target slots, while the blue spots are the start points for a specific task T (i.e., “task 11”). The ego vehicle pose is determined by ( 24 ).
Fig. 8 : A human driver is performing a parking maneuver.
Sensor
Count
Resolution
FOV
RGB Camera
4
400×300 px
100∘
Depth Camera
4
400×300 px
100∘
IMU
1
Accelerometer (3-axis),
Gyroscope (3-axis), Compass
GNSS
1
Provides Latitude, Longitude
Vehicle State
–
Position, Orientation, Speed,
Appendix
TABLE III : Consolidated Sensor Specifications
Fig. 9 : Distribution Map of Parking Start Point. All target parking slots are normalized to the center of the image. It can be seen that for parking in one side, there are 20 start points (2 columns by 10 rows), accompanied by a certain amount of jitter.
Fig. 10 : Overview of data distributions for core metrics, grouped by the number of gear shifts. From left to right: Average Speed, Slice Length, and Trajectory Length.
Category
Avg. Speed
Slice Len.
Traj. Len.
Count
(km/h)
(frames)
(m)
1 -shot
Unconstrained
25
2 -shot
[2.90,6.70]
[101,210]
[12.8,25.2]
111
3 -shot
[2.79,5.70]
[125,219]
[14.0,22.5]
120
4 -shot
[2.83,6.00]
[144,280]
[17.0,31.0]
306
Total
562
Appendix
TABLE IV : Data Filtering Thresholds and Final Task Distribution
Automated parking requires generating complete and executable trajectories in highly constrained spaces with low tolerance for goal pose error. Existing end-to-end parking methods struggle to jointly achieve inference efficiency, trajectory quality, and precise endpoint alignment, while conventional imitation objectives provide limited supervision on structured deviations from expert maneuver geometry. We propose DriftParking, a one-step trajectory generation framework that reconstructs the drifting-field paradigm for high-precision conditional trajectory generation. Specifically, we replace distribution-level attraction with conditional one-to-one attraction toward the paired expert trajectory, introduce expert-centered constructive repulsion, and adaptively attenuate repulsion near convergence. We further formulate trajectory generation in an endpoint-residual space by decomposing each trajectory into a start-to-goal baseline and a learnable residual, turning endpoint alignment into a representation-level structural constraint on the supervision target while providing a structured space for repulsive supervision. DriftParking achieves state-of-the-art performance across all evaluation metrics. Closed-loop on-vehicle experiments across diverse parking scenarios further show a 97% parking success rate, demonstrating strong zero-shot generalization.
End-to-end autonomous parking has emerged as a critical task within the realm of autonomous driving. However, existing methods suffer from black-box characteristics, lacking high-level semantic understanding and interpretability, which impedes the realization of seamless long-distance autonomous parking from the road to the target spot. To address these limitations, we propose ParkingTransformer, a novel framework that leverages multi-view perception and the scene understanding capability of Large Language Models (LLMs). By combining trajectory queries with LLMs implicit state features, our method interacts directly with historical information and raw sensor data to output planning trajectories, eliminating the need for dense Bird's-View (BEV) representations. To compensate for the inadequate spatial reasoning ability of LLMs, we introduce 3D positional encoding to explicitly inject spatial geometric awareness. Furthermore, a fixed-window streaming mechanism is designed for historical information processing, significantly improving long-term temporal processing efficiency and inference speed. Additionally, a coarse-to-fine decoding strategy is employed to progressively enhance trajectory precision. Extensive closed-loop experiments are conducted on the CARLA simulator and real-world vehicle platforms. The results demonstrate that our method achieves a driving score of 61.32 in CARLA simulator and an average success rate of 88.70% in real-world experiments, validating the feasibility and effectiveness of the proposed algorithms.
Hauteng Wu, Xu Li, Dong Kong +4
School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China · School of Electronic and Information Engineering, Tongji University, Shanghai, 201804, China · College of Transportation, Shandong University of Science and Technology, Qingdao 266590, China +1
Autonomous parking remains a critical yet challenging task in intelligent driving systems, particularly within constrained urban environments where maneuvering space is limited and precise control is essential. While recent advances in end-to-end learning have shown great promise, the lack of high-quality, structured datasets tailored for parking scenarios remains a significant bottleneck.To address this gap, we present ParkingScenes, a comprehensive multimodal dataset specifically designed for end-to-end autonomous parking in simulated scenes. Built on the CARLA simulator, ParkingScenes features structured parking trajectories generated by a Hybrid A* planner and a Model Predictive Controller (MPC), providing accurate and reproducible supervision signals. The dataset includes 16 reverse-in and 6 parallel parking scenarios, each executed under two pedestrian conditions (present and absent), resulting in 704 structured episodes and approximately 105000 frames. Each scenario is repeated 16 times to ensure consistent coverage. Each frame contains synchronized data from four RGB cameras, four depth sensors, vehicle motion states, and Bird's-Eye View (BEV) representations, enabling rich multimodal fusion and context-aware learning. To demonstrate the utility of our dataset, we compare models trained on ParkingScenes with those trained on unstructured, manually collected simulation data under identical conditions. Results show significant improvements in performance, underscoring the effectiveness of structured supervision for robust and accurate parking policy learning. By releasing both the dataset and the collection framework, ParkingScenes establishes a scalable and reproducible benchmark for advancing learning-based autonomous parking systems. The dataset and collection framework will be released at: https://github.com/haonan-ai/ParkingScenes
Haonan Chen, Kaiwen Xiao, Bin Tian +1
Institute of Automation · Chinese Academy of Sciences · Beijing, China +3