SPIRo: An AI Based Origami Inspired Soft Robot with Multidimensional Locomotion and Multimodal Data Analysis for Infrastructure Assessments

Current methods of pipe infrastructure assessment are heavily dependent on humans, making them dangerous, expensive, inefficient, and often missed early leak detection. With the high amount of toxic gas being released into the atmosphere and the millions of dollars worth of damage caused by gas leaks every year, there is an urgent need to develop a new method of inspecting gas pipes. This paper presents the first AI and origami based infrastructure assessment robot, SPIRo. It is able to move in a worm-like motion at a speed of 12 mm/s through individually actuated McKibben artificial muscle actuators, which offer enough strength to allow the robot to extend and contract. The robot utilizes curved magnetic grippers to attach to any magnetic pipe surface. To accurately detect gas leaks, deep learning of thermal images and gas composition data is applied. The initial testing of these deep learning algorithms, the convolutional neural network (CNN) for thermal images and artificial neural network (ANN) for gas sensor data, has achieved an accuracy of approximately 92% and 93% respectively for data collected in a simulated test environment. All of these characteristics combined enable the functionality of an autonomous infrastructure assessment robot.