Intelligent collision management in dynamic environments for human-centered robots

Thanks to rapid breakthroughs on robotics, their historical deployment in industrial setups, their current extensions to warehouses, and the highly anticipated deployment of autonomous vehicles on our streets, self-guided mobile robots equiped with manipulators or varied payloads are paving their way into unstructured and dynamic environments such as cities, hospitals, human-populated work areas, and facilities of all types. As a result, intentional or unintentional contact between humans, objects and these robots is bound to occur increasingly more often. In this context, a major focus of this thesis is on unintentional collisions, where a straight goal is to eliminate injury from users and passerby’s via realtime sensing and control systems. A less obvious focus is to combine collision response with tools from motion planning in order to produce intelligent safety behaviors that ensure the safety of multiple people or objects. Yet, an even more challenging problem is to anticipate future collisions between objects external to the robot and have the robot intervene to prevent imminent accidents. In this dissertation, we study all of these sophisticated flavors of collision reaction and intervention. The posit here is that no matter how hard we try, collisions will always happen, and therefore we need to confront and study them as a central topic both during navigation or dexterous manipulation. We investigate in-depth multiple key and interesting topics related to collisions and safety of mobile robots and robotic manipulators operating in human environments. We show that simple sensor architectures can reconstruct sophisticated external force information including location, direction, and magnitude over the whole body of some types of robots. We devise technologies to quickly sense and react to unexpected collisions as fast as possible during navigation. We investigate robots colliding against walls in difficult tilted terrains and quickly figuring out new practical motion plans. We study methods to recognize intentional contacts from humans and use them as a non-verbal communication medium. We study fusing sensor data from contact sensors and time of flight laser sensors to reason about the multiplicity of contacts on a robot from human users. We investigate statistical problems like the probability that an externally moving object collides against a human. We then devise novel motion planning and control algorithms to stop the impending collisions using any part of a robot’s upper humanoid body. Such behaviors constitutes some of the most advanced collision mitigation and intervention techniques we have seen in the academic communities. Overall we deeply investigate collisions from many perspectives and develop techniques that borrow and contribute to the areas of mechatronic design, sensor processing, feedback controls, motion planning, and probabilistic reasoning methods. The result of this study is a set of key experiments and guidelines to deal with collisions in mobile robots and robotic manipulators. This study aims at influencing future studies on field operations of robots and accelerate the employment of advanced robots in our daily environments without compromising our safety.Mechanical Engineerin