Mutually Adaptive Shared Control between Human Operator and Autonomy in Ground Vehicles

Abstract

Semi-autonomous driving can leverage the benefits of self-driving vehicle technologies today, even if the latter may still be limited in scope and reliability for fully autonomous driving. In semi-autonomous driving, the human driver collaborates with autonomy to control the vehicle. Among the various frameworks to enable this collaboration, this thesis focuses on the haptic shared control framework, because it allows for a continuous negotiation between the two agents, which has been shown to be beneficial over discrete transition schemes.

An essential question for shared control is how to allocate the control authority between the two agents. The literature presents static and adaptive schemes, the latter allowing for better driving performance. To date, several vehicle and human factors have been considered for adaptation, but a crucial human factor has been omitted: workload. Workload represents the availability of the driver's mental resources and is important especially when the driver engages in multiple tasks as expected in a semi-autonomous driving setting. Therefore, the overarching goal of this thesis is to develop adaptive haptic shared control schemes for semi-autonomous driving that take workload into account and evaluate their benefits over non-adaptive schemes.

To achieve this goal, this thesis develops workload-adaptive control schemes at two levels. The first adaptation developed is at the control consolidation level, i.e., after the autonomy commands are generated and during their blending with the human's commands. The developed scheme modulates autonomy commands as a function of driver workload. Human subject studies are used to assess the performance of the developed schemes, where the subjects' goal in the driving task is to track a path with the help of autonomy and complete a surveillance task simultaneously. Results from the human subject studies reveal that compared with the non-adaptive control consolidation, the developed adaptive control consolidation scheme can achieve similar driving performance with less control effort from the human under the conditions when there is a minor disagreement between agents. Under conditions where there are significant disagreements between the two agents in the perception of the path center, the adaptive control consolidation can reduce workload, path tracking error, and control effort while increasing trust.

The second adaptation is developed within the autonomy. It enables adaptation of the autonomy's control commands independent of how they are blended with the human's control command. Results show that, compared with non-adaptive cases with different maximum speed limits, it can balance the driving performance, duration, emergency maneuvering performance, human's control effort under emergency, and self-reported workload while reducing the control effort in the lane-keeping task.

Finally, simultaneous adaptation at both levels is also evaluated through human subject tests. Results show that when both adaptive schemes are used, they can balance driving performance, duration, and workload with less steering effort from the human when compared with the non-adaptive cases. Compared to using adaptive autonomy alone, adding the adaptive control consolidation further reduces the control effort under emergency conditions and the driving duration. Compared to using the adaptive control consolidation alone, adding the adaptive autonomy reduces the control effort in the lane-keeping task and achieves a more robust driving performance.

Therefore, this dissertation makes the following original contributions: 1. A workload-adaptive control consolidation scheme and its performance evaluation. 2. A workload-adaptive autonomy and its performance evaluation. 3. Evaluation when adaptation happens in both control consolidation and autonomy.