Haptic Steering Interfaces for Semi-Autonomous Vehicles

Abstract

Autonomous vehicles are predicted to significantly improve transportation quality by reducing traffic congestion, fuel expenditure and road accidents. However, until autonomous vehicles are reliable in all scenarios, human drivers will be asked to supervise automation behavior and intervene in automated driving when deemed necessary. Retaining the human driver in a strictly supervisory role, however, may make the driver complacent and reduce driver's situation awareness and driving skills which ironically, can further compromise the driver’s ability to intervene in safety-critical scenarios. Such issues can be alleviated by designing a human-automation interface that keeps the driver in-the-loop through constant interaction with automation and continuous feedback of automation's actions. This dissertation evaluates the utility of haptic feedback at the steering interface for enhancing driver awareness and enabling continuous human-automation interaction and performance improvement in semi-autonomous vehicles.

In the first part of this dissertation, I investigate a driving scheme called Haptic Shared Control (HSC) in which the human driver and automation system share the steering control by simultaneously acting at the steering interface with finite mechanical impedances. I hypothesize that HSC can mitigate the human factors issues associated with semi-autonomous driving by allowing the human driver to continuously interact with automation and receive feedback about automation action. To test this hypothesis, I present two driving simulator experiments that are focused on the evaluation of HSC with respect to existing driving schemes during induced human and automation faults.

In the first experiment, I compare obstacle avoidance performance of HSC with two existing control sharing schemes that support instantaneous transfers of control authority between human and automation. The results indicate that HSC outperforms both schemes in terms of obstacle avoidance, maneuvering efficiency, and driver engagement. In the second experiment, I consider emergency scenarios where I compare two HSC designs that provide high and low control authority to automation and an existing paradigm that decouples the driver input from the tires during collision avoidance. Results show that decoupling the driver invokes out-of-the-loop issues and misleads drivers to believe that they are in control. I also discover a `fault protection tradeoff': as the control authority provided to one agent increases, the protection against that agent's faults provided by the other agent reduces.

In the second part of this dissertation, I focus on the problem of estimating haptic feedback from the road, or the road feedback. Road feedback is critical to making the driver aware of the state of the vehicle and road conditions, and its estimates are used in a variety of driver assist systems. However, conventional estimators only estimate road feedback on flat roads. To overcome this issue, I develop three estimators that enable road feedback estimation on uneven roads. I test and compare the performance of the three estimators by performing driving experiments on uneven roads such as road slopes and cleats.

In the final part of this dissertation, I shift focus from physical human-automation interaction to human-human interaction. I present the evidence from the literature demonstrating that haptic feedback improves the performance of two humans physically collaborating on a shared task. I develop a control-theoretic model for haptic communication that can describe the mechanism by which haptic interaction facilitates performance improvement. The model creates a promising means to transfer the obtained insights to design robots or automation systems that can collaborate more efficiently with humans.