Sebastian Zanlongo is a PhD candidate and Department of Energy Fellow studying Computer Science with a specialization in motion planning and multi-robot coordination. He has developed novel scheduling and planning algorithms at G2, Inc., multi-robot behavior strategies at Los Alamos National Laboratories, and machine learning techniques for anomaly detection at Sandia National Laboratories. Sebastian is currently working on methods for multi-robot adaptive sampling in radioactive environments at the Applied Research Center and with Dr. Leonardo Bobadilla at the MoRA lab.
There is an abundance of nuclear material in the world that requires constant monitoring. Semi and fully autonomous robots can navigate in areas that are dangerous or inaccessible for humans. However, current systems are often ad-hoc and brittle, unable to scale to large numbers of robots. Moreover, robot movements are hindered by the confined spaces of these old facilities, increasing the difficulty of monitoring these locations. We therefore propose three main thrusts to solve issues that are hindering progress in environmental monitoring, and provide accompanying solutions.
- Robot task allocation in complex environments: Given partially unknown environments, heterogenous robots, and tasks with varying requirements and deadlines, how do we best allocate tasks to robots? We provide an on-line solution that allows multiple robots to visit locations throughout an environment and increase the number of tasks that are completed ahead of their deadlines.
- Robot policy generation given operator constraints: Robots often require operator supervision when in dangerous environments. However, we cannot scale a 1:1 operator-to-robot ratio, we must instead have multiple robots assigned to a single operator while avoiding conflicts. How do we allocate operator attention across multiple robots? Our solution uses a novel geometric approach to represent the possible robot configurations, while sampling techniques yield a policy.
- Informative path planning in constrained environments: Given a large environment and limited movement, it can be difficult to effectively sample to form a model. What is the best method to select sampling locations to simultaneously reduce costs while still providing an accurate regression? We plan to use informative path planning techniques to localize areas of interest, and accompany this with a path-planning strategy that can cope with the constrained environment.