Longitudinal Jerk Estimation for Identification of Driver Intention

摘要

We address the problem of estimating online the longitudinal jerk desired by a human driver piloting a car. This estimation is relevant in the context of suitable identification of driver intentions within modern Advanced Driver Assistance Systems (ADAS) such as the co-driver scheme proposed by some of the authors. The proposed architecture is based on suitably combining a Kalman filter with a scaling technique peculiar of the context of "high-gain" observers. The scaling is appealing because it allows for an easy tuning of the trade-off between phase lag and sensitivity to noise of the resulting estimate. Additionally, we show that using engine-related experimental measurements available in the CAN bus, it is possible to provide a more reliable estimate of the driver-intended jerk, especially in the presence of gear changes. The proposed scheme shows very desirable results on experimental data from a track test, also when compared to a brute force approach based on a mere kinematic model.