Reliability of Multi-Sensor Fusion for Next Generation Cars and Trucks 2014-01-0718

Cars and trucks today are getting fitted with a large number of sensors in an effort to improve safety, comfort, fuel economy and emissions. The revenue from the automotive sensors market, driven by intense global competition and regulation, is expected to double over the next decade, while the size of the automotive sensors market, over the same period, is expected to triple.

The field of sensor-fusion is highly multi-disciplinary, making use of technics from artificial intelligence, pattern recognition, digital signal processing, control theory, and statistical estimation. Sensor-fusion strategies based on probability theory, evidence theory, fuzzy theory, and possibility theory are being explored in different industries, e.g., defense, robotics, automotive, etc. The majority of sensor-fusion operators are based on optimistic assumptions about reliability of the information generated by the sensors. However, many or all sensors in a fused sensor system may exhibit substantially different reliability levels over the life of a vehicle, and it is necessary to account for this variation/degradation to avoid any decrease in performance of the fusion results. Many of the automotive sensor applications are tightly tied with safety critical actuation systems which need to perform control actions reliably over the entire life of the car or the truck. The safe and reliable control of the vehicles requires that as different sensors change in their performance capability and reliability, over the course of the life of the vehicle, the sensor fusion strategies take the sensor system changes into account adequately in order to maintain an acceptable level of reliability of the fusion results.

A large part of sensor-fusion literature relies on the strong assumption that the sensors producing the fused data operate according to a predetermined characteristic at all times. Applications and investigations related to monitoring of reliability of a number of fused sensors, as the performance of the sensors varies with time, including sensor failure, have been reported relatively infrequently. This paper presents an overview of sensor-fusion investigations in the automotive arena and the relatively small body of work concerned with reliability of sensor-fusion. The paper also highlights the role of Evidence theory in forming the foundation for sensor-fusion reliability developments that could help car and truck makers realistically take advantage of the growing volume of sensors in their products.