Automated Valet Parking
Date: 2025-08-28
Location: Cranfield MUEAVI test track
Keywords: Automated Valet Parking, Novel Perception unit, LiDAR, 3D ultrasonic sensor
Person Collecting Data: Hari Vijayakumar, Research Fellow in Navigation for Connected and Automated Mobility, Automotive, Energy and Photonics Theme
Contact: hari.vijayakumar@cranfield.ac.uk
Principal Investigator: Marco Cecotti, Lecturer, Automotive, Energy and Photonics Theme
Contact: M.cecotti@cranfield.ac.uk
Funding Source: UK Research and Innovation, grant number 10059986.
Methodological information: 
These two videos show the results of the project Drive By Sound, in which an automated valet parking system was tested using LiDAR and a novel ultrasonic sensor. A mapping of a parking Area around AIRC was initially done using LiDAR, and this has been translated and rotated to the MUEAVI test track to ensure Safety. As the testing was to evaluate the performance of the 3D ultrasonic sensor at low speed, the maximum speed during the operation has been limited to 1 m/s. The experiment vehicle has been fitted with a drive-by wire unit supplied by an American Company, Signal, which will be used to control, throttle, brake and steering actuators:
* Initially, the map of the proving ground configurable area has been marked and obtained using the Lanelet2 toolbox. Then, the map of the parking area has been obtained using a process called georeference SLAM process with LiDAR and OXTS RT1003. The vehicle localises itself within the map using the OXTS RT1003, which has been equipped with differential GPS correction, resulting in a position accuracy of 10cm. The mapped parking area is then translated and rotated to the proving ground drivable area for further testing.
* To incorporate live detections into the offline Map and to save computation, the 3D map is thus converted to 2D by slicing along the vertical axis (z in our case) and then converted to an obstacle cost map. To this extent, the obstacles are inflated by the vehicle's width, and a lethal cost is assigned to this. Furthermore, an exponentially decaying cost is applied beyond the inflation radius. Hence, the vehicle will avoid following too closely to the detected obstacles.
* A hybrid A* based path planner is implemented to generate a path from the start to the destination, and while replanning around the obstacles. A velocity planner that takes into account the rate of change of acceleration is used to maximise the passenger comfort
* When confronted with an obstacle, the vehicle will stop in front of the obstacle and replan around it. Once the path is replanned, the vehicle waits until the operator sends a prompt to the system confirming that the path is safe and that they can continue with the rest of the journey.
* The longitudinal control implementation was based on PID, and the Stanley controller was used as the lateral tracker.
* The video (DBW_ValetParkingLiDAR) is based on LiDAR perception, where the obstacles can be detected well before, as the range of the sensor is ~70m. Hence, when an obstacle is detected, the vehicle has enough time to decelerate smoothly without affecting passenger comfort. The video also shows the driver's point of view, feeds from the front and side cameras, and the vehicle model in a mapped environment, enabling tracking of the vehicle's real-time behaviour.
* The video (DBW_ValetParkingCalyo_UltrasonicSensor) is based on 3D ultrasonic sensor perception, where the obstacles are detected much later compared to LiDAR, however, still giving enough time to decelerate without affecting the passenger comfort. The optimal range of the ultrasonic sensor was found to be around ~12-15m. The video also shows the driver's point of view, feeds from the front and side cameras, and the vehicle model in a mapped environment, enabling tracking of the vehicle's real-time behaviour.

Sharing and access information: 
* Yet to be published in a paper
* Citation: Vijayakumar, Harikrishnan and Cecotti, Marco (2026). Driven By Sound - Automated Valet Parking. Cranfield Online Research Data (CORD). Media. https://doi.org/10.57996/cran.ceres-2843
Licence: CC-BY-ND