Robotic Perception in Extreme Environments

点击翻译按钮获取中文摘要

Rapid advances in artificial intelligence are driving the adoption of robotics and automation in transport and logistics, providing new solutions to passenger transport, last-mile delivery, and automated warehouses. These systems carry a high level of risk to public safety and it is our duty as robotics engineers to design the safest possible systems. Since environmental perception systems are essential to any autonomous robotic system it is imperative to avoid errors in these systems. This is all the more important in extreme environments (in challenging weather, underwater environments, caves, etc.), where sensors (cameras, sonars, etc.) are subject to high levels of noise and frequently produce erroneous measurements. Tragedies involving Lion Air's Boeing 737 Max and Tesla are sad examples of the consequences of erroneous sensor and perception systems failures. The mitigation of spurious measurements has been studied in perceptual tasks such as the recording of point clouds required for remote sensing (i.e. using LiDAR) or the alignment of 3D shapes. There is an urgent public safety need to address these issues, especially with the emergence of transport and logistics robotic systems, such as autonomous flying or driving vehicles. In this research program, we will address how to autonomously reduce the adverse effects of erroneous measurements on robotic perception systems in extreme environments. There is no comprehensive solution to perception failures in robotics, especially in presence of erroneous measurements, which could be a showstopper to the deployment of robots in society. With more robust perception, robots will be safer and will be able to venture into increasingly hostile environments. Eventually, provable robustness could lead to certifiable systems, which would allow their introduction in safety-critical applications. In particular, we are interested in rapid mapping, localization, and navigation in limited-access and communication-restricted areas (e.g. space exploration, mining, etc.), where perception failures could lead to material loss and danger to human life. To achieve this goal we request a Boston Dynamics Spot, an all-terrain agile robot with an extensive sensor suite, that we plan to deploy in mapping missions in extreme environments such as caves and mountains in collaboration with space agencies in North America and Europe.