Autonomous Robots for Scientific Monitoring of Marine Environments

是什么阻碍了加拿大水体的机器人环境监测实现完全自动化？答案是双重的。首先，今天在该领域的环境数据收集受到空间和时间的限制，因为它(a)完全由人类科学家执行，因此耗时，稀疏，昂贵且规模有限，或者(b)它是由静态传感器完成的，这些传感器稀疏地覆盖感兴趣的体积并产生大量数据，其中只有一小部分对科学家来说是重要的。其次，也是最关键的一点，机器人忽略了专家科学家真正需要和重视的数据类型。机器人不知道下一步哪里值得导航，也不知道哪里值得关注。本项目涉及的研究活动和支持他们的设备以一种整体的方式解决了这两个问题。

What is preventing robotic environmental monitoring of Canadian water bodies from being fully automated? The answer is twofold. First, today's environmental data collection in the field is spatially and temporally constrained because it is (a) either performed entirely by human scientists and is therefore time-consuming, sparse, expensive, and of limited scale, or (b) it is done by static sensors which sparsely cover the volume of interest and yield a lot of data, only small portions of which is important to the scientist. Second, and most crucial, robots are oblivious to the type of data that expert scientists really need and value. Robots do not know where is worth navigating next or where to pay attention to. The research activities involved in this project and the equipment that will support them address these two questions in a holistic way. Autonomous robot exploration informed by the human scientist provides a fruitful middle ground between these two desiderata, and allows faster exploration, in broader areas, while at the same time being attentive to scientifically relevant data of high value. Using this equipment we will demonstrate efficient collaborative robot exploration between an autonomous aerial vehicle and an autonomous boat, equipped with water quality sensors and samplers, as well as cameras (multi-spectral, RGB, and thermal), and bathymetry and mapping sensors. Deployments will happen in challenging marine environments in Lake Ontario, pushing the envelope of what has been demonstrated in the robotics literature thus far in terms of systems and algorithms development. More specifically, this equipment will support the development and validation of computer vision, machine learning, and control algorithms to address challenging autonomous sampling, planning under uncertainty, and human-robot interaction problems in outdoor natural environments. It will enable sensing, perception, and autonomous robot navigation in the field.