Robotic teamwork and collaboration

点击翻译按钮获取中文摘要

In the short term, I seek to develop algorithms to address problem of developing autonomous active observers: systems that can automatically select and collect information regarding the world around them. I will do this big focussing, in particular, on aspects of this problem where a target agent that wants its activities to be documented is being followed (or preceded) by one or more documentarians with which it seeks to cooperate, but without explicit communication or explicitly shared state. The observation or recording of events in the world is critical to our understanding of our environment and our management of it, be it a sequence of events in a power plant, the interactions of a biologist in an ecosystem that in being evaluated, the activities of an automated system at work, or the activities of the police for during a major event. This grant will address this problem on the theoretical front by building computational models that explain under what circumstances we can compute actions can be planned to allow a robot to follow a sequence of events associated with another moving agent, even allowing the use of multiple observers and multiple targets. This theoretical development will be coupled with the production of robotic systems that employ this theory to operate in real time using video input data to document other robots, people and/or their environment. As robotics and machine sensing have started to mature, the study of interactions between robots, and between robots and their environment has become the lynchpin of practical applications. My research program addresses multi- agent cooperative interaction specifically in the context of information-seeking tasks. These can be informally envisioned as the creation of principles supporting the development of intelligent observers, or more colloquially a class of “robotic video documentarians.” That is, the development of automated systems that can document evolving events in the world. To fully accomplish this kind of task, a robot must develop a plan that gets it in the right vantage points at the right times by predicting the movement of the dynamically changing subject(s), obstacles and occlusions. In the last 6 years my lab has made progress on both the theoretical constraints that govern robotic systems, as well as the experimental development and evaluation of robotic systems that walk, swim, and fly, with variants that operate indoors or outdoors, and even interact with one another across these disparate domains. This work has resulted in the last 6 years in roughly 50 refereed publications, and that base has allowed me to acquire experience solving multi-robot interaction problems in both theory, i simulation and in practice. To address the problem of developing robotic documentarians, I will begin by developing algorithms to address three inter-related aspects of the larger problem: determination of what visual events to observe and when, computation of optimal motion plans to allow intermittent observation of these visual events without explicit communication, and computing plans for the recovery of visual contact if the agents have become separated. Each of these algorithmic stages will then find embodiment in functioning laboratory systems where their performance in practice can be evaluated in controlled circumstances and compared with theoretical predictions and bounds.