← All publications

SICE Annual Conference, 2024

Angle-Aware Full 3D Coverage Control with ADMM-based Dynamic Assignment of Charging Stations

Zhiyuan Lu, Muhammad Hanif, Takeshi Hatanaka

pp. 1119–1124.

Part of Angle-Aware Coverage Control for 3D Map Reconstruction

Paper
Angle-Aware Full 3D Coverage Control with ADMM-based Dynamic Assignment of Charging Stations

Angle-aware coverage control assumes the drones stay in the air. Batteries say otherwise: a field large enough to be worth reconstructing takes longer to cover than a drone can fly, so charging has to be part of the control problem rather than an interruption to it.

Enforcing a charging constraint is not new. What hurts is which station a drone returns to. Fix the drone-to-station pairing in advance and a drone can end up flying the length of the field to reach “its” station, wasting exactly the flight time the constraint was meant to protect.

Angle-aware full 3D coverage control: a drone with gimbal camera observing points in the target field
The underlying problem — full 3D angle-aware coverage with a steerable camera, so each drone's state is position plus two gimbal angles.

Approach

This paper assigns stations dynamically, through the alternating direction method of multipliers (ADMM), and folds that assignment into the constraint-based controller from the camera-rotation work. The QP each drone solves carries four barrier constraints at once:

  • coverage performance — the decay rate of the objective stays above γ/n
  • battery — the drone can always still reach its assigned station
  • gimbal pitch stays within its mechanical limits
  • no collision with the nearest drone

Each has a slack variable, so the problem stays feasible even when the specifications conflict.

Simulation

Built on ROS 2 Humble with JAX for JIT compilation and GPU acceleration, four drones modelled on the DJI Mavic 3E, four charging stations, and takeoff/landing delays added to make the simulator honest about the cost of going down to charge.

Six snapshots of the ROS 2 simulation showing four drones, their view cones, charging stations as green cylinders, and the field colour-coded by importance
Red is high importance, green is covered. Green cylinders are charging stations — their height and colour show the battery level of the drone currently assigned, and the white line marks the assignment. A reassignment happens between 6 s and 7 s; by 55 s all four drones are on the ground charging; by 400 s the field is nearly green.

Against a fixed one-to-one assignment, the dynamic version drives the coverage objective down faster, for the reason it was designed to: nobody crosses the field to charge.