Issue 59 Uncrewed Systems Technology Dec/Jan 2025 Thunder Wasp UAV | Embedded computing tech | SeaTrac USV | Intergeo | UAVE 120 cc four-stroke | Launch & recovery | Magazino UGV | DroneX | Knightsbridge K5 security robot

21 Data for Level 5 Decision-making software in autonomous vehicles relies on information from ADAS sensors. Reliable data on their performance under diverse road, traffic and weather conditions is essential to build the trust that enables authorities to certify vehicles to use Level 5 autonomy. “There are quadrillions and quadrillions of bytes of machine-learning data that need to be gathered before we can even think about Level 5,” Oakes says. Ensuring the data from these sensors remains reliable throughout the service life of the vehicle fleet, even after accident repairs and demanding recalibrations, is a major challenge for the industry, placing vehicle technicians in maintenance and repair shops in the frontline. They are also in the frontline of consumer education. “The consumer needs to know what is on their vehicle, why it’s there, and how the interaction with ADAS technology affects their driving. Even more than a ‘check engine’ light, it is important to have an ADAS warning checked out immediately because it is all about safety,” Oakes says. Some devices, such as proximity sensors mounted in bumpers and cameras in door mirrors, are easy to damage in even minor knocks and scrapes, while others that feed sensor-fusion software, such as bioptic and stereoscopic cameras, radar, Lidar and night-vision cameras, are more robust but not invulnerable. Therefore, vehicle technicians need to understand their operating principles and recalibration requirements. Even routine maintenance procedures on ADAS-equipped vehicles demand a higher degree of specific knowledge. “A simple, four-wheel alignment may now require the vehicle’s steering angle sensor (SAS) to be reset. If the technician does not research the vehicle’s equipment before service – and the manufacturer requires an additional ADAS recalibration like that – it may go unchecked.” Recalibration issues Regardless of the technical education of the consumer and technician, the biggest challenge in the repair of ADASequipped vehicles is the size of the bill, as sensor replacement and recalibration are expensive, often to the point of being prohibitive to the average consumer. “I recently saw a vehicle – fresh from the body shop – with an obvious camera issue: one outside mirror had a camera, the other did not,” Oakes says. “When I asked, the repair manager told me that the owner had replaced his own mirror before the accident as the cost for a camera-equipped unit was approximately $750, with installation and ADAS recalibration costing extra.” While recalibration requirements and complexity vary between sensors, vehicles and installations, these differences count far less than the technician’s skills, and their development and upkeep through education and training. Oakes argues that the installation and resetting of a vehicle’s sensor technology is similar to any other muscle-memory task, unlike diagnosis and programming. These systems are highly softwareintensive, so dealing with bugs and keeping them up to date is a constant state of focus for Tier 1 suppliers and automotive manufacturers working together to gather as much operational data as possible. However, some adjustments need to be made to improve data quality and tighten clustering – the process of grouping similar sets of data points to make it easier for machine-learning algorithms to interpret and act on the information. Pam Oakes | In conversation The consumer needs to know what is on their vehicle, why it’s there, and how the interaction with ADAS technology affects their driving Uncrewed Systems Technology | December/January 2025 Before cars like Audi’s Grandsphere concept can operate at autonomy Level 5, a vast amount of precise, accurate ADAS data has to be clustered and processed to ensure the trustworthiness of sensor data and algorithmic decisions (Image courtesy of Audi)

RkJQdWJsaXNoZXIy MjI2Mzk4