Improving Human-Centered Robotic Safety Filters Through Slot Car Racing

Abstract

Our current world is a coexistence between humans and technology. This does not always lead to good outcomes. In this thesis, we explore safety filters on semi-autonomous technologies. The purpose of these filters is to protect and assist human operators of these systems. However, in practice, there are several instances where the safety filters cause automation surprises, alarming and confusing the human operator. In some instances, this has led to extremely unsafe scenarios. In this thesis we demonstrate how to implement two working safety filters on a slot car system through hardware and software implementations, thereby improving human compatibility. This thesis implements two working safety filters on a miniature racetrack system. The monitoring of the track and the car’s condition is the same for both, yet the fallback strategies vary. We set out to determine which is more helpful for the human operator. The safety filters’ intervention mechanism is on the hardware level, instead of the software level, by applying a voltage to the track that controls the speed of the car as it traverses a lap on the racetrack. By creating a hardware-level intervention mechanism, we were able to focus on the human-machine interface, exploring human-centered safety filters. Besides ensuring safety, these filters’ goal is to increase the transparency of the automation technologies and thereby lead to fewer automation surprises for the human operator. Through experimentation with human subjects, we were able to test how human operators interacted and responded to both types of safety filters, and determine which features they found more user-friendly, intuitive, and collaborative to their intent of operating the system. This was a central goal of the thesis: determine how to build safety filters that increase safety and seamlessly integrate with the human operator experience. Improving compatibility between humans and automated technologies, which are increasingly present in all industries around us. Through experimentation we found great success with one filter, leading to 100% compatibility with our test users. All participants tested with this safety filter re- ported they would use the filter in real-life scenarios given the opportunity. The second filter, which had a more aggressive style led to 50% of the users reporting they would use the filter in a real-life scenario. Nonetheless, both safety filters increased the efficiency of laps traversed around the racetrack and decreased the failure rates of user subjects. Therefore accomplishing the ultimate goal of this thesis. Ultimately, we have proven that we can implement two different safety filters on this testbed system and are able to, through hardware interventions, create more human-centered and human-operator-friendly safety filters.