RADRON: Cooperative Localization of Ionizing Radiation Sources by MAVs with Compton Cameras

TL;DR

This work tackles radiation-source localization in hazardous environments by deploying a decentralized swarm of lightweight MAVs, each equipped with a single-detector Compton camera. It introduces a multi-agent extension of Compton-cone data fusion using an onboard, low-cost Kalman-filter-based estimator and a novel decentralized flocking controller to maximize sensor baselines. The approach yields dramatic improvements in initialization speed and tracking accuracy, enabling real-time localization and tracking of moving radiation sources with a small, agile swarm. Realistic simulations and open-field experiments demonstrate the swarm’s ability to quickly initialize, maintain encirclement around a source, and track moving sources, reducing human exposure and expanding operational capabilities in GNSS-denied or obstructed environments.

Abstract

We present a novel approach to localizing radioactive material by cooperating Micro Aerial Vehicles (MAVs). Our approach utilizes a state-of-the-art single-detector Compton camera as a highly sensitive, yet miniature detector of ionizing radiation. The detector's exceptionally low weight (40 g) opens up new possibilities of radiation detection by a team of cooperating agile MAVs. We propose a new fundamental concept of fusing the Compton camera measurements to estimate the position of the radiation source in real time even from extremely sparse measurements. The data readout and processing are performed directly onboard and the results are used in a dynamic feedback to drive the motion of the vehicles. The MAVs are stabilized in a tightly cooperating swarm to maximize the information gained by the Compton cameras, rapidly locate the radiation source, and even track a moving radiation source.

Figures (9)

• Figure 1: Three cooperating MAV localizing and tracking a moving radiation source. Each MAV (red) is equipped with a miniature Compton camera and an onboard computer. The ionizing radiation source (a sample of Cesium-137) is carried by a quadruped robot (yellow).
• Figure 2: Cone reconstruction in a single-detector Compton camera. A high-energy photon (blue) is emitted by a source $\mathbf{x}$. The Compton scattering occurs on the first contact with the detector and the recoiled electron is immediately absorbed. The scattered photon (red) passes through the detector until it is also absorbed. Energies $E_{\lambda'}$, $E_{e'}$ are measured directly by the pixel matrix bonded to the rear face of the detector. The active pixels pinpoint the $x, y$ coordinates of both events $\mathbf{c}_{e'}, \mathbf{c}_{\lambda'}$. The difference in $z$ is computed from the time delay between the two detection events and the velocity of event propagation (obtained in calibration). The scattering angle $\theta$ is computed using (\ref{['eqn:compton']}). Finally, a set of all possible origins of the high-energy photon is reconstructed. This set forms the surface of a cone $\mathbb{C}$.
• Figure 3: An example with 2 MAV using heterogeneous positioning systems. Each drone $i,j$ estimates its position and orientation relative to the origin $\mathbf{O}_i, \mathbf{O}_j$, respectively. The Compton cones $\mathbb{C}_i, \mathbb{C}_j$ are also projected in the respective coordinate frames. Under (A2), the position of the radiation source $\mathbf{x}$ can be estimated in both coordinate frames using all available measurements.
• Figure 4: Measurement correction done by projecting the latest hypothesis $\mathbf{x}_{k-1}$ onto the surface of a cone $\mathbb{C}$ to get a corrected hypothesis $\mathbf{x}_k$.
• Figure 5: A top-down view of the swarm's self-organization during the encirclement of the hypothesis. Here shown for two MAV with starting positions $\mathbf{m}_{1}, \mathbf{m}_{2}$. The signed angle from $\mathbf{m}_{1}$ to the nearest other MAV is $\theta_1$. To steer the swarm towards a uniform spacing angle $\theta^{*}$, a bias $\beta$ is applied to the start of the generated trajectory $\mathbf{t}_1$. The bias is applied in the orientation opposite to $\theta$. The first point of the desired trajectory $\mathbf{t}_1[0]$ does not correspond to the position of the MAV $\mathbf{m}_1$. However, the discontinuities in the desired trajectory are already tackled at a lower level of the control architecture baca2018model. This allows the swarm controller to retain simplicity and generate new trajectories at a high rate.