Thanks to advancements in sensing technology and ICT, autonomous driving is promising to revolutionize transport and mobility. Unfortunately, current systems do not adapt well to unexpectedly complex and/or dynamic situations and environments. This limits the value of autonomous driving at best and causes bodily harm at worst.

The pursuit of improved robustness and overall performance has resulted in a trend towards sensor fusion, i.e. to overcome individual sensor limitations by combining the strengths of multiple sensors. This project will research a new sensor-fusion concept called adaptive (cooperative) sensor fusion.

Cooperative sensor fusion is a paradigm that allows sensors to communicate between themselves weak evidence at low-data rates. We will expand the theory of this paradigm through expanding its grounding in estimation theory. This will enable us to better understand and exploit the value of said weak evidence, which will be built upon to “steer” or “adapt” the way sensors acquire data (adaptive sensing) and/or process said data (adaptive processing). This paradigm will enable us to optimize the perception process at the system-level, allowing to optimize not just robustness and performance, but also system-level latency, energy use, interference caused to other agents…. Note that in this adaptive paradigm feedback loops are created: adaptive sensing creates new and adapted weak evidence, which in turns adapts the sensing. Initially these loops will be created in a very conservative way so as not to cause undesired effects. Nonetheless, these loops form an interesting research challenge in their own right and will lead to further research and, when done right, will bring a new level of performance and robustness to automotive perception systems.