Collaborative Research: CISE-MSI: DP: RI: Towards Scalable, Resilient and Robust Foraging with Heterogeneous Robot Swarms

The project aims to create a highly efficient, resilient, and robust heterogeneous robot swarm system composed of UAVs (unmanned aerial vehicles) and UGVs (unmanned ground vehicles) for foraging tasks in a large unknown environment such as search and rescue, agriculture harvesting, and space exploration. There are three key challenges that hinder the efficiency of foraging robot swarms in practice.

Firstly, while homogeneous ground robot swarms are efficient, they face limitations in Tsearching for multiple targets or resources in a large area due to limited sensing and mobility capabilities. Secondly, machine learning models that determine robot behavior are traditionally trained on a central server, which is not scalable when dealing with heterogeneous robots with varying configurations.

Finally, sensor malfunctions and adversarial attacks are likely to occur in robot swarms and can lead to cascading effects that reduce the robustness and resilience of the swarm. This project leverages a promising interdisciplinary approach across robotics, machine learning, and cybersecurity to achieve a scalable, robust, and resilient foraging robot swarm with heterogeneous robots.

The proposed education plan aims to create new curricula in related fields at both the undergraduate and graduate levels and engage K-12 students in research through various initiatives such as robot expo, summer research camps, and workshops. It will also promote the participation of Hispanic students in research, education, and outreach activities in the South Texas region.

The proposed research explores the design of heterogeneous robot swarm systems, multiple shortest path planning, federated learning (FL), and spatiotemporal data anomaly detection. The research plan consists of three research thrusts. 1) Decentralized multiple shortest-route planning algorithm will be developed for on-demand UAV sensing. This algorithm will enable UAVs to efficiently sense and explore interesting locations along the shortest routes, allowing for timely data collection while conserving energy resources. 2) Decentralized federated learning (FL) algorithms will be developed to support privacy-preserving collaborative model training in the heterogeneous robot swarm.

The proposed algorithms allow for the customization of a model for each robot, making the swarm more scalable and resilient as the number of robots increases. 3) The sub-trajectory discord based early anomaly detection module to early detect robot failure/attacks in the resource constraint environment, preventing the cascading effect in the entire swarm. The study of anomaly detection will ensure the resilient and robust of the model learned through FL.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.