Skip to content
Atikul Islam Munna logo Home
Back to projects

Research · 2026

resQswarm: Secure Search and Rescue Drone Swarm

Manuscript in preparation

Abstract

resQswarm studies how a small drone swarm can keep searching when its coordinator is lost or attacked. Five simulated drones search a disaster area under a policy trained with multi-agent reinforcement learning. One drone acts as the mother: it fuses the shared search map, assigns sectors and sends heartbeats.

If the mother is destroyed, stops responding or is compromised, a child drone takes over through a signed succession protocol, and this repeats down to the last drone. Every drone flies on a custom cascaded flight controller and carries a security layer against GPS spoofing, RF jamming and firmware attacks.

Approach

  • Two simulation levels: training in GPU-vectorised VMAS and evaluation in full PyBullet quadrotor physics, with one policy that runs in both.
  • A custom cascaded flight controller with attitude, rate and altitude loops, validated against a reference PID controller under gusts and waypoint jumps.
  • Signed swarm messaging with replay protection and attestation, plus attacker models for GPS spoofing, jamming and compromised firmware.
  • Sensor models with datasheet-based noise parameters and a lawnmower sector sweep as the search baseline.

Current status

The flight controller, simulation worlds and sensor models are implemented. Swarm training and the attack experiments are in progress.

Research areas

  • Multi-Agent RL
  • Drone Swarms
  • Robotics
  • Cyber-Physical Security
  • Simulation