Autonomous mobile rescue robot for casualty extraction

Abstract

The development of robotic systems for search and rescue (SAR) operations offers a great promise in potentially minimising the danger to rescue workers and improving their efficiency. Rescue robots also offer the flexibility to easily scale up the rescue fleet and enable rescue workers to carry out large-scale operations. Even though a large number of research studies have been conducted in the field of search and rescue robotics, substantially fewer works have focused on the development of rescue robots capable of performing physical rescue interventions, including loading and transporting victims (a.k.a. casualties) to a safe zone, a process known as casualty extraction. The main objective of this thesis is to develop a proof-of-concept mobile rescue robot capable of rescuing a person lying on the ground with a safe and reliable casualty extraction procedure and a high degree of autonomy. To achieve this objective, three key important aspects of developing a reliable autonomous casualty extraction robot are investigated in this thesis: (1) Mechanical system to enable the robot to navigate through a real-world environment and effectively perform the desired casualty extraction procedure while ensuring it satisfies the required safety measures. Two design iterations and proof-of-concept implementations of robot prototypes called ResQbot 1.0 and ResQbot 2.0 are presented, including the proposed methods for safely loading a victim from the ground onto the robot. (2) Perception system to provide a reliable casualty detection method in the presence of visual disruptions. The investigation includes a novel ground projected point cloud (GPPC) method for processing the point cloud data input, a novel ResQNNet architecture for deep-learning-based casualty detection, and implementation of the sim-to-real learning method with specific data augmentation strategies. (3) Control system to enable the proposed mobile rescue robot to safely manoeuvre around and approach the target victim as part of the autonomous casualty extraction procedure. A novel hierarchical decomposed-objective model predictive control (HiDO-MPC) method is proposed and evaluated in simulated environments and real-robot setups using the ResQbot 1.0 platform. The key findings and outcomes from this thesis demonstrate that the proposed proof-of-concept prototypes are capable of safe and effective execution of the proposed casualty extraction procedure with a high degree of autonomy. The proposed implementation achieves reliable casualty detection and robust control performance while dealing with the uncertainty present in the conducted real-world experiments.