Range-Based Multi-Robot Integrity Monitoring Against Cyberattacks and Faults: An Anchor-Free Approach

2 min read

Vijay and colleagues use inter-robot range measurements (UWB-style) as a cross-check on each robot’s own position estimate. The monitor looks for the sparsest set of per-robot position errors that explains the mismatch between measured ranges and the ranges implied by the estimates. That gives a detection signal (overall integrity), an identification signal (per-robot integrity against a threshold), and an estimate of how far off each affected robot is. No robot is assumed correct, unlike anchor-based methods. The nonlinear problem is solved by sequential convex programming with ADMM so it can be distributed, and the solver is cold-started when the network topology changes.

Findings

  • Numerical simulation with 20 UAVs, 6 of them GNSS-spoofed. Error reconstruction converged for three of four noise and penalty settings and diverged with high noise (0.05) and a high penalty (1.25).
  • With a two-phase attack (3 robots each phase) and 100 Monte Carlo trials per setting, warm-starting with penalty 0.75 diverged after the second phase. Cold-starting was much more reliable at large penalties.
  • Mixed-reality demo with 4 real Crazyflie 2.0 drones and 3 PX4 software-in-the-loop drones: a constant GNSS offset on one real and one virtual drone pushed their integrity measures over the threshold.

Limits

  • Assumes the error vector is sparse (few affected robots) and that range noise is bounded. The threshold is set by hand from the noise bounds.
  • In the demo, ranges were emulated from motion capture and the monitor ran centrally on a laptop. The distributed hardware version is future work (authors).
  • It reconstructs and flags errors but does not isolate robots or change the mission.
  • An attack and a fault that produce the same position error look identical, and the paper treats them as one class.

Relevance

A geometric check against GPS spoofing in aerial swarms; gil-2017-spoof-resilient attacks a similar problem with radio-signal fingerprints. It fits runtime monitoring and robot security, and shows the familiar pattern: cyberattack or fault, the monitor cannot tell which. The same group’s coverage work is pant-2025-resilient-coverage.