Robust Parking Path Planning with Error-Adaptive Sampling under Perception Uncertainty.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Automated parking systems rely on path planners to guide vehicles to vacant spots.
• Perception systems introduce uncertainty (e.g., sensor noise, algorithmic errors), risking collisions.
• Generating safe paths under perception uncertainty is a significant challenge.

Purpose of the Study:

• To develop a robust parking path planning algorithm that mitigates risks from perception uncertainty.
• To integrate perception error estimation and adaptation into the path planning process.
• To ensure safe and accurate vehicle positioning within parking spaces despite detection inaccuracies.

Main Methods:

• Proposed a robust parking path planning method combining error-adaptive sampling and utility-based decision-making.
• Error-adaptive sampling generates diverse path candidates considering potential detection errors.
• Utility-based method selects the optimal path based on a defined utility function under uncertainty.

Main Results:

• The integrated algorithm continuously generates adaptable paths that account for detection errors.
• Demonstrated enhanced safety by ensuring vehicles reach the true parking space position and orientation.
• Successfully addressed the challenge of planning paths from erroneous parking space detections.

Conclusions:

• The proposed robust parking path planning algorithm effectively manages perception uncertainty.
• This approach enhances the safety and reliability of automated parking systems.
• Future work could explore advanced sensor fusion techniques to further reduce perception uncertainty.