Safe and Efficient Robot Learning in Human-Centric Environments

点击翻译按钮获取中文摘要

Robotics is poised to revolutionize several aspects of our society in the coming years from the way we move people and things around to healthcare to manufacturing. In particular, mobile robotics technology has matured to the point that self-driving cars and unmanned aerial vehicles (a.k.a., drones) are being actively pursued by industry, from small startups to the world's technology giants. While some products will come to market in the next few years, they will not be fully autonomous, meaning they will not operate under all conditions that a human operator can handle. To achieve this, a great deal more research is need. Real applications of mobile robots are challenging because the operating environment (e.g., our roadways, our skies, big warehouses) can be large, cluttered, dynamic, and unknown/uncertain. This means that the key autonomy functions, which include localization, mapping, planning, and control, must address these challenges. Within robotics, we have barely scratched the surface when it comes to making systems that can operate at a human level. This program of research focuses on expanding the operational envelope of robots to complex and human-centric environments (i.e., environments built for and used by humans such as our roads networks, city centers, buildings, etc). The proposal focuses on techniques that enable robots to act safely and make dependable decisions in dynamic and uncertain situations with multiple actors. Prior knowledge is combined with data to achieve fast adaptation to new situations. In particular, we will address (i) fast adaptation to dynamic disturbances, (ii) data-based motion planning, (iii) performance guarantees in uncertain environments, and (iv) interactive scenarios. These are all important elements of developing autonomy for use in human-centric environments, which tend to be less predictable ahead of time and require robots to operate near other dynamic actors. The work as part of this proposal will include fundamental methodological advances, the implementation of novel algorithms in software, and thorough experimental testing in realistic scenarios. We will embed our new autonomy features into the full autonomy stack comprising all the key functions needed to operate mobile robots through a dynamic and uncertain world. As we do so, we will work closely with industrial partners to transfer our technology to real-world applications such as warehouse robots, self-driving cars, and drone delivery systems. Canada is rapidly growing its capabilities in these areas and we hope to make big contributions in terms of the technology and people needed to shape the future.