PDC Sensor for Autonomous Parking - Sensor Fusion and Real-Time Control for Precision Parking Maneuvers
This technical article explores the sensor fusion and real-time control aspects of autonomous parking using PDC sensors. It covers the integration of ultrasonic data with other sensors (camera, radar), the trajectory tracking control, the obstacle avoidance logic, and the fail-safe mechanisms for reliable self-parking.
Autonomous parking requires seamless sensor fusion between PDC sensors, surround-view cameras, and ultrasonic radar (if present). The PDC sensors provide precise distance measurements at close range (0.2-4 m), while cameras provide visual information for recognizing parking space markings and potential objects, and radar provides longer-range detection for approach. The fusion algorithm uses a probabilistic occupancy grid approach, where each cell in the grid is assigned a probability of being occupied based on the measurements from all sensors. The ultrasonic data is weighted more heavily for cells near the vehicle (within 2 m), while camera data is weighted for identifying lines and markings. The fused grid is used for path planning and obstacle avoidance. The sensor fusion also helps in handling situations where one sensor type is degraded (e.g., camera blinded by sunlight, ultrasonic affected by dirt). The system uses a data association algorithm to ensure that measurements from different sensors refer to the same physical objects.
PDC Sensor
Trajectory tracking control during autonomous parking uses the PDC sensor data to maintain the vehicle on the planned path with sub-decimeter accuracy. The control loop compares the actual distances to the obstacles with the expected distances from the path plan. Any error is corrected by adjusting the steering angle using a pure pursuit or Stanley controller. The controller gains are tuned to balance responsiveness with smoothness. The PDC sensors also provide feedback on the vehicle's lateral position relative to the parking spot boundaries, enabling fine adjustments during the final positioning. The system uses the distance measurements to trigger a stopping command when the vehicle reaches the desired final position (e.g., when the rear bumper is at a certain distance from the wall). The control is executed at a high rate (50 Hz) to ensure precise maneuvering.
Obstacle avoidance logic is built into the control algorithm. During parking, the PDC sensors continuously monitor the surroundings for any new obstacles that may appear (e.g., a pedestrian walking behind the vehicle). If an obstacle is detected within the safety envelope (e.g., 0.5 m), the system pauses the maneuver and issues a warning. If the obstacle moves closer, the system may abort the parking and return to a safe position. The obstacle avoidance is based on a dynamic window approach, where the control commands that lead to a collision are rejected. The system also uses the rate of change of distance to predict if a moving obstacle will cross the path, adjusting the maneuver accordingly. The sensors' fast update rate (50 Hz) is crucial for reacting to dynamic obstacles.
Fail-safe mechanisms ensure the system can handle sensor failures or unexpected events. The autonomous parking system has a redundant sensor architecture: if one PDC sensor fails, the system can still use the remaining sensors, though the coverage may be reduced. The system continuously monitors the sensor outputs for plausibility (e.g., distance not exceeding maximum range, consistent with other sensors). If a sensor fault is detected, the system automatically downgrades the parking mode (e.g., from fully autonomous to semi-autonomous) and alerts the driver. Additionally, the system has a time-out mechanism: if the parking maneuver takes longer than a specified time (e.g., 3 minutes), the system aborts and requests driver intervention. The fail-safe design is critical for safety and is validated through extensive testing.
The future of autonomous parking with PDC sensors involves the use of higher-frequency sensors (e.g., 60 kHz) for better resolution, and the integration of AI-based decision-making for more complex parking scenarios (e.g., parallel parking on curved streets). The sensors are also being developed with built-in signal processing to provide object classification (e.g., wall vs. pedestrian) to enhance the decision-making. The autonomous parking system is evolving toward a fully automated valet parking system, where the vehicle can find a parking spot, park, and later retrieve itself without driver involvement. This requires even more robust and accurate ultrasonic sensing, making PDC sensors an indispensable component of the future mobility ecosystem.
