Driver Steering Torque Estimation for Driver Behavior Modeling During Steering Assistance

In autonomous driving and advanced driver-assistance systems, establishing an accurate driver model that reproduces human steering behavior is essential for achieving smooth, cooperative control between human and machine.

While driver steering torque information is crucial for constructing such models, it is difficult to measure directly using onboard sensors during actual driving.

In this study, we formulated a state estimation problem based on experimental data, treating the unobservable driver torque as an unknown disturbance. To mitigate the phase lag (estimation delay) typically encountered in conventional disturbance observer approaches, we proposed a smoothing estimation method using a Kalman smoother. Verification results showed a significant reduction in Root Mean Square Error, realizing highly accurate offline driver torque estimation.