Haptics based collaborative steering control in automated driving environments

A vehicle's steering is a particular system in that it is exposed to individual subjective reviews based on criteria that are hard to assess quantitatively. Haptic design of such systems is a prime concern that has been at the center of industrial development for the past 30 years. Despite the recent rise of advanced driving assistance systems (ADAS) and automated driving (AD) functions, technical and legal challenges still bind the human driver to the driving task and further emphasize haptic design as a communication channel for collaboration between human driver and automation. This thesis contributes to the development of a comprehensive collaborative steering control framework that enables a consistent haptic design in situations where the driver and the automation share steering tasks. The critical concern is how to control automation to enable rich interaction with the human driver. This control framework, inspired by human-human haptic collaboration, relies on three major functions: interaction enables the driver and the automation to steer together, arbitration sets how the automation interacts with the driver, and inclusion allows the automation to adapt its trajectory by assimilating the driver intent. Arbitration is an attractive research topic not only in shared steering but also in general human-robot joint tasks. However, estimation of the human motor control is challenging due to the limited sensors in typical steering systems. To facilitate vehicle implementation, a simplified approach for the estimation of the driver sensorimotor control and the adaptive control of the automation impedance using these estimates was first implemented in a prototype vehicle to validate the effectiveness of the overall control framework. The quantitative assessment of several drivers suggests that the proposed framework has the capability to lead to smoother collaborative behavior with less effort than conventional approaches. In addition, the simplifications made on the arbitration for the sake of practicality (i.e. a non-optimal arbitration rule and reduced accuracy estimation of the driver sensorimotor) have been reconsidered by relaxing the constraint on computational load. First, the arbitration strategy was optimized by using a nonlinear model predictive control. This approach enables the arbitration of a wide range of interaction types (e.g. assistance, education, co-activity, collaboration, and competition) by minimizing a cost function. The proposed arbitration rule can represent all interaction types by merely changing the weights of the cost function. This is effective especially in joint tasks such as driving applications, where the type of interaction is supposed to change dynamically depending on the traffic condition. Second, an interacting multiple model filter has been implemented to estimate the driver sensorimotor control. This approach enables simultaneous estimation of the driver goal and impedance in real-time, without the need for additional sensors or training for data-driven methods. The proposed framework has the potential to extend conventional ADAS or AD functions, which are operated typically in a discontinuous way, towards continuous shared operation, which results in a more consistent steering feel of the driver. Furthermore, this general framework is not limited to steering tasks, but can be applied to any other collaborative tasks between human and robot.