Human factors of semi-autonomous robots for urban search and rescue

During major disasters or other emergencies, Urban Search and Rescue (USAR) teams are responsible for extricating casualties safely from collapsed urban structures. The rescue work is dangerous due to possible further collapse, fire, dust or electricity hazards. Sometimes the necessary precautions and checks can last several hours before rescuers are safe to start the search for survivors. Remote controlled rescue robots provide the opportunity to support human rescuers to search the site for trapped casualties while they remain in a safe place. The research reported in this thesis aimed to understand how robot behaviour and interface design can be applied to utilise the benefits of robot autonomy and how to inform future human-robot collaborative systems. The data was analysed in the context of USAR missions when using semi-autonomous remote controlled robot systems. The research focussed on the influence of robot feedback, robot reliability, task complexity, and transparency. The influence of these factors on trust, workload, and performance was examined. The overall goal of the research was to make the life of rescuers safer and enhance their performance to help others in distress. Data obtained from the studies conducted for this thesis showed that semi-autonomous robot reliability is still the most dominant factor influencing trust, workload, and team performance. A robot with explanatory feedback was perceived as more competent, more efficient and less malfunctioning. The explanatory feedback was perceived as a clearer type of communication compared to concise robot feedback. Higher levels of robot transparency were perceived as more trustworthy. However, single items on the trust questionnaire were manipulated and further investigation is necessary. However, neither explanatory feedback from the robot nor robot transparency, increased team performance or mediated workload levels. Task complexity mainly influenced human-robot team performance and the participants’ control allocation strategy. Participants allowed the robot to find more targets and missed more robot errors in the high complexity conditions compared to the low task complexity conditions. Participants found more targets manually in the low complexity tasks. In addition, the research showed that recording the observed robot performance (the performance of the robot that was witnessed by the participant) can help to identify the cause of contradicting results: participants might not have noticed some of the robots mistakes and therefore they were not able to distinguish between the robot reliability levels. Furthermore, the research provided a foundation of knowledge regarding the real world application of USAR in the United Kingdom. This included collecting knowledge via an autoethnographic approach about working processes, command structures, currently used technical equipment, and attitudes of rescuers towards robots. Also, recommendations about robot behaviour and interface design were collected throughout the research. However, recommendations made in the thesis include consideration of the overall outcome (mission performance) and the perceived usefulness of the system in order to support the uptake of the technology in real world applications. In addition, autonomous features might not be appropriate in all USAR applications. When semi-autonomous robot trials were compared to entirely manual operation, only the robot with an average of 97% reliability significantly increased the team performance and reduced the time needed to complete the USAR scenario compared to the manually operated robot. Unfortunately, such high robot success levels do not exist to date. This research has contributed to our understanding of the factors influencing human-robot collaboration in USAR operations, and provided guidance for the next generation of autonomous robots