Road-Shoulder Scanning Using Multi-Sensor Kalman Filter for Minimum Risk Maneuver 2021-01-0867

Determining availability of the road shoulder is a key to the minimum risk maneuver when autonomous or semi-autonomous vehicles encounter problems and autonomous mode is unavailable. Failure to detect available road shoulder in a fast manner may place autonomous or semi-autonomous vehicles in hazardous conditions for extended time, while false detection of unavailable road shoulder may lead to crashes or run off the road, which is dangerous too. On the other hand, determining availability of the shoulder is challenging problem given state of art perception signals. None of the perception signals are robust enough to correctly detect shoulder availability for all circumstances. This paper presents an algorithm of road-shoulder assessment by utilizing multi-sensor Kalman filter. Three sources of available road shoulder length are from radar tracker, camera, and drivable corridors, which is detection algorithm using radar inputs. By cross-checking these three sources, different combination of sources might be selected and be feed into a Kalman filter, which is an application of classic multi-sensor Kalman filter. In order to evaluate the effectiveness of the proposed algorithm, the algorithm is implemented in a dSpace MicroAutoBox and 1 hour and 20 min driving is conducted. The front camera is used to record the driving environment and the ground-truth of availability of road shoulder is done by viewing the recorded video and manually labelling. From the testing results, the performance of detection available road shoulder has been greatly improved compared with camera-only method.