Adaptive Data Collection for Rapid Evaluation of New Plant Varieties

A looming question that must be solved before robotic plant phenotyping capabilities can have significant impact to crop improvement programs is scalability. Current field phenotyping systems rely on exhaustive coverage of a breeding experiment. This works well in relatively small proof-of-concept trials, but will fail as the size of the field gets larger. This is especially true for ground-based platforms, which can collect higher quality data than aerial platforms, but have much lower coverage rates. This project is investigating two algorithmic approaches to scaling up the rate at which ground robots can effectively cover a field: informative path planning and multi-robot coverage control. In one result, we consider the problem of online robotic sampling in environmental monitoring tasks where the goal is to collect k best samples from n sequentially occurring measurements. Using the information theoretic criterion, we present an online submodular algorithm for stream-based sample selection with a provable performance bound. We also consider the problem of online environmental sampling and modeling for multi-robot sensor coverage, where a team of robots is deployed over the workspace in order to optimize the overall sensing performance. For this problem, we propose a new approach with mixture of locally learned Gaussian Processes for collective model learning and an information-theoretic criterion for simultaneous adaptive sampling in multi-robot coverage. This approach demonstrates a better generalization of the environment modeling and improved performance of coverage without assuming the density function is known a priori.