Selecting Spots by Explicitly Predicting Intention from Motion History Improves Performance in Autonomous Parking
Long Kiu Chung, David Isele, Faizan M. Tariq, Sangjae Bae, Shreyas Kousik, Jovin D'sa
AI summary
Problem
Autonomous valet parking requires selecting safe, socially acceptable parking spots, but existing methods struggle with ambiguous long-term goals, lack realistic simulation assumptions like occlusion and reactive agents, and rely on imperfect trajectory predictions.
Approach
The pipeline reconstructs semantic bird's-eye view maps for surrounding vehicles using probabilistic belief maps, then explicitly predicts their parking intentions from motion history to guide spot selection and trajectory planning.
Key results
- Novel BEV reconstruction via belief maps enables intention prediction for unobserved agents
- Intention-conditioned spot selection improves prediction accuracy and social acceptance
- Cubic Bézier trajectory prediction balances computation time, accuracy, and task completion
- Outperforms trajectory-based and end-to-end implicit baselines in realistic simulations
Why it matters
Provides a practical, modular framework for safe and socially acceptable autonomous parking in complex, occluded environments, advancing real-world AVP deployment.
Abstract
In many applications of social navigation, existing works have shown that predicting and reasoning about human intentions can help robotic agents make safer and more socially acceptable decisions. In this work, we study this problem for autonomous valet parking (AVP), where an autonomous vehicle ego agent must drop off its passengers, explore the parking lot, find a parking spot, negotiate for the spot with other vehicles, and park in the spot without human supervision. Specifically, we propose an AVP pipeline that selects parking spots by explicitly predicting where other agents are going to park from their motion history using learned models and probabilistic belief maps. To test this pipeline, we build a simulation environ- ment with reactive agents and realistic modeling assumptions on the ego agent, such as occlusion-aware observations, and imperfect trajectory prediction. Simulation experiments show that our proposed method outperforms existing works that infer intentions from future predicted motion or embed them implicitly in end-to-end models, yielding better results in pre- diction accuracy, social acceptance, and task completion. Our key insight is that, in parking, where driving regulations are more lax, explicit intention prediction is crucial for reasoning about diverse and ambiguous long-term goals, which cannot be reliably inferred from short-term motion prediction alone, but can be effectively learned from motion history. Website: https://sites.google.com/view/chung2026selecting.