MODELLING HUMAN-AUTOMATION INTERACTIONS IN A HAPTIC SHARED CONTROL FRAMEWORK
Analytics
52 views ◎36 downloads ⇓
Abstract
This thesis is focused on modeling the interaction between a human driver and anautomation system . While numerous companies and academic groups are pushingto develop autonomous vehicles with the aim of freeing up attention for drivers andimproving safety on the road; barriers remain for deployment of fully autonomousvehicles, including technical, legal, and social barriers. The challenge of meetinghuman capabilities for sensing, perceiving, and predicting the environment on theroad is formidable, even more challenging is the hand-off problem of how to achievea smooth transition of control authority between a human driver and automationsystem. Combining the best capacities of a human driver with the speed, accuracy,and tirelessness of automation will require a shared control framework that is intuitivefor the driver.In this thesis, we explore the interaction of a human driver and an automationsystem in a haptic shared control framework. In a haptic shared control framework,the human driver and automation system both act on the steering wheel, exertingcontrol on the vehicle but also communicating with each other using haptic cues andsignals. Both the human driver and automation system act with limited impedance:the human by biomechanics and the automation system by design, with the use ofproportional control.In this thesis, the interaction between the two agents (i.e., the human driver andautomation system) are modeled using a game-theoretic approach. The human andautomation system are both modeled with a similar structure. Specifically, the humanmodel consists of a higher-level controller representing his cognitive controller,as well as a lower-level controller representing his biomechanics. Similarly, the automation system is modeled with a higher-level controller (AI) as well as a lower-levelimpedance controller. Since the human and automation dynamics can adaptivelychange by modulating their impedance (lower-level controller), the higher-level controllerof the human and automation system is modeled using an adaptive modelpredictive controller.When there exist two controllers, that is, the human driver and the automation system,there is the possibility that their objectives in terms of target paths are conflicting,and the corresponding control actions are thereby non-cooperative. To explorethe interaction between the driver and the automation system under such conditions,two-games equilibrium strategies known as non-cooperative Nash and Stackelberg arederived, and some simulation results related to these equilibrium types are presentedand discussed. The Nash paradigm represents a scheme where both agents act asleaders (i.e., leader-leader) in performing a task. On the other hand, the Stackelbergparadigm presents a case where one agent acts as a leader, while the other acts as afollower. It is shown that for the same impedance, the Stackelberg solution achievesa possibly better path-following performance than the corresponding Nash solution.