Urban search and rescue (USAR) has been a significant application in robotics because the deployment of rescue robots in these life-risking tasks can help reduce the personal risks of human rescue workers and increase the speed of emergency response. Furthermore, the incorporation of multi-robot rescue teams consisting of a group of heterogeneous rescue robots is advantageous since it allows shorter robot deployment times, provides situational awareness from multiple locations, and introduces redundancy and fault-tolerance to this time-critical mission. While promising, the control of multi-robot rescue teams is a challenging task. In particular, decisions need to be made regarding which rescue tasks should be executed and by which robot in a specific scenario.

This thesis focuses on the development of semi-autonomous control architectures for multi-robot USAR teams in order to address the challenges in human-robot coordination and multi-robot cooperation in USAR applications. A learning-based semi-autonomous control architecture is developed for multi-robot teams in USAR environments, allowing a team of rescue robots to learn and make decisions regarding which rescue tasks need to be carried out at a given time. The task sharing between rescue robots and human operators can reduce the stress and mental workload of the operators, while allowing rescue robots to benefit from a human operator’s experience and knowledge. Furthermore, the cooperation between different rescue robots enables fast exploration of cluttered USAR scenes by minimizing overlap or repeated revisits of an explored area, as well as improve the reliability of the overall search and rescue mission.

This thesis further addresses the high-level task allocation challenge for a large heterogeneous multi-robot team. A hierarchical supervisory control architecture is proposed allowing effective task sharing amongst the multi-robot team based on the dimension, mobility and sensing capabilities of individual robots. The developed approach can also handle the changes to team size due to robot failures and the need for robot repair during a USAR mission. The developed multi-robot task allocation approach allows a single operator to control a large team of heterogeneous rescue robots in USAR applications. Simulations and experiments in simulated USAR environments verified the performance of the proposed supervisory controller.