NARRO2W SPACE- Networked Adversarially-Resilient Reconfigurable Operations in ObstacleWorlds for Safe Planning and Attack-tolerant Control and Estimation (Panagou)
Abstract
The proposed project will explore fundamental conditions and develop algorithms under which networked agents reconfigure themselves in time and space, and maintain secure, and at the same time less conservative, communication structures in the presence of safety constraints. To this end, we will investigate adaptation algorithms for adjusting the level of network (graph) robustness on-the-fly, so that the level of resilience changes based on the updated knowledge about the environment, the adversarial agents and their attacks. The core technical development is the concept of Adversarially-Resilient Tunable Control Barrier Functions (A-RT CBFs), whose super-level sets will encode the safety and resilience requirements that should be met, or relaxed in a principled manner, along the network trajectories. Such developments would go beyond standard fault detection or resilient consensus, as the environment itself poses challenges that are related with 1) how to safely plan the motion of a team of agents in narrow spaces based on geometric constraints and the updated knowledge about the environment (adversarial attacks and attacker identification), and 2) how to relax these constraints in a principled and systematic way, if needed, so that a minimum tolerance to adversarial attacks can be guaranteed with concurrent satisfaction of safety constraints. The research outcomes will provide novel metrics of resilience and safety in dynamic, confined environments, along with algorithmic developments that realize computationally-efficient and improved performance for multi-agent networks in narrow spaces. The algorithms will be evaluated in open-source simulators (AirSim, Gazebo) and experiments with ground and aerial robots using open-source software in the PI’s research lab in the Ford Robotics Building and in M-Air, the netted outdoors facility at the University of Michigan, North Campus.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310163
Entities
People
- Dimitra Panagou
Organizations
- Air Force Office of Scientific Research
- Board of Regents of the University of Michigan
- United States Air Force