Evaluation of a Sensor Fusion Algorithm on a Real-Time Processor

Abstract:
Heterogeneous systems offer the possibility to achieve a high performance while requiring little energy and therefore generating only little waste heat. Using them inside a vehicle makes it feasible to implement many algorithms for sensor processing utilized by autonomous cars. Managing the distribution of different algorithms among the available heterogeneous processor cores is difficult. Especially in areas which require additional safety mechanisms, the selection of cores is crucial. In this paper an evaluation is presented regarding the real-time capability of a deterministic real-time processor running in lockstep mode to increase the safety. The test program consists of a sensor fusion algorithm that combines the results of sensor processing algorithms executed on unsafe but fast processing units. This approach allows to use algorithms with high hardware performance requirements while still maintaining redundancy by moving important features like the sensor fusion to slow but reliable processors.