A system for single human supervision of multiple robots in urban search and rescue

• Main Author: Wong, Choon Yue.
• Other Authors: Seet Gim Lee, Gerald
• Format: Theses and Dissertations
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Mechanical engineering::Mechatronics; DRNTU::Engineering::Mechanical engineering::Robots
• Online Access: http://hdl.handle.net/10356/53910
• Institution: Nanyang Technological University

The deployment of multiple-robots in dangerous applications like urban search and rescue (USAR) requires the robots to be supervised, possibly by only one human. This gives rise to single-human multiple-robot systems (SHMRSs). Supervising robots can increase workload and limit the supervisor’s situation awareness. Robots in a group should autonomously coordinate their actions to mitigate these issues. When designing autonomous coordination, it is important to consider the limited communication and computation resources on robots. This study has adopted a holistic approach that considers the limitations in human cognition as well as in robot communication and computation to design a solution for robot deployment in a SHMRS. Gaps have been identified in the existing knowledge. The literature has suggested that computation for coordination should be centralized and performed by a single apex robot. This reduces the communication and computation overheads for subordinate robots. However, the effects of adopting a centralized organization structure for the robot group in a SHMRS are not well documented. The interaction effects between the centralized organization structure and the number of robots deployed (group size) are also unclear. In addition, there is a need to demonstrate that involving computationally simpler subordinate robots in coordination would not necessarily limit coordination sophistication. While each robot in a distributed robot group communicates with all group members, a subordinate robot in a centralized group communicates only with the apex. Researchers have thus assumed that centralized robot groups require less communication between robots during coordination. This assumption has not been verified. This study has proposed a solution based on equipping the SHMRS with an explicitly communicating, strongly centralized robot group with a co-located apex. The solution also specifies a direct communication channel between the supervisor and each robot. Such channels maintain communication between supervisor and robots even after apex incapacitation. A control solution was developed to address the knowledge gaps. The control solution was a SHMRS that also featured explicitly communicating robots but used a distributed robot group. The SHMRSs based on both solutions have been implemented in the real world. An experiment was conducted with a simulated USAR scenario. Eight participants supervised robot groups of four possible configurations based on the centralized and distributed structures as well as group sizes of two and three robots. Performance measures included the number of victims found, the number of robots incapacitated, supervisor workload, supervisor situation awareness and the volume of inter-robot communication. While the small sample size resulted in a less-than-ideal level of statistical significance, the experiment served to demonstrate the system and provided an insight regarding these performance attributes using the developed system. A proof-of-concept demonstration of the SHMRS showed the robots simultaneously deployed to search for victims while supervised by a single human. It was possible to produce coordination that was as sophisticated as that achieved with a distributed robot group despite using subordinate robots in the centralized robot group. While the supervision of a centralized robot group induced more workload due to the participants’ motivation to ensure apex-robot survival, that workload promoted their awareness of the robots’ situation. This improved awareness contributed to a reduced rate of robot incapacitation when centralized robot groups were supervised. The limitations of the proposed solution were that the added cognitive workload when supervising the centralized robot group degraded task performance. Also, supervising three instead of two robots for centralized robot groups resulted in a greater increase in supervision difficulty. Unlike robots in the distributed robot group, subordinate robots required the apex to inform them of coordination results. The centralized robot group thus required more communication when coordinating in the experiment scenario.