Automated Valet Parking

Abstract

This open-source project implements a complete motion-planning pipeline for automated valet parking in constrained, unstructured environments. It starts with a Hybrid A* path and progressively produces a dynamically feasible trajectory through path optimization, cubic interpolation, velocity planning, and optimal control.

The implementation is designed for tight parking maneuvers, with collision checking and optimization-based refinement used to improve path quality and trajectory feasibility. Benchmark scenarios are represented as parking maps with initial and goal vehicle poses, obstacles, and their vertices.

Keywords

Automated Valet Parking
Hybrid A*
Path Optimization
Trajectory Planning
Optimal Control

Methods

Vertical end-to-end Automated Valet Parking framework from scenario modelling and Hybrid A-star search through path smoothing, interpolation, velocity planning, and IPOPT-based optimal control. — End-to-end automated valet parking framework: scenario modelling and Hybrid A* initial-path search, path smoothing and interpolation, then speed planning and IPOPT-based optimal control.

1. Initial Path Generation

Hybrid A* searches for a collision-free coarse path that connects the initial and goal poses in a parking map with geometric obstacles.

2. Path Smoothing and Interpolation

The coarse path is optimized and then interpolated with cubic curves to obtain a smooth geometric reference suitable for vehicle motion.

3. Velocity Planning and Optimal Control

A velocity profile is planned along the reference path, and an optimization problem is solved with IPOPT to generate the final dynamically feasible parking trajectory.

Experiments

The following benchmark cases are reproduced from the project repository. Each row pairs the final static trajectory visualization with its corresponding animated parking process.