# Observability-driven Assignment of Heterogeneous Sensors for Multi-Target Tracking

## TL;DR

This paper presents a matroid-theoretic greedy algorithm for assigning heterogeneous sensors to multi-target tracking tasks, optimizing tracking quality with proven approximation guarantees and demonstrated effectiveness in simulations.

## Contribution

It introduces a novel assignment algorithm leveraging matroid theory for heterogeneous sensors, with proven approximation bounds and practical efficiency.

## Key findings

- Algorithm achieves near-optimal tracking performance.
- Proven approximation bounds of 1/3 and 1/2 for different functions.
- Effective in long-term multi-target tracking simulations.

## Abstract

This paper addresses the challenge of assigning heterogeneous sensors (i.e., robots with varying sensing capabilities) for multi-target tracking. We classify robots into two categories: (1) sufficient sensing robots, equipped with range and bearing sensors, capable of independently tracking targets, and (2) limited sensing robots, which are equipped with only range or bearing sensors and need to at least form a pair to collaboratively track a target. Our objective is to optimize tracking quality by minimizing uncertainty in target state estimation through efficient robot-to-target assignment. By leveraging matroid theory, we propose a greedy assignment algorithm that dynamically allocates robots to targets to maximize tracking quality. The algorithm guarantees constant-factor approximation bounds of 1/3 for arbitrary tracking quality functions and 1/2 for submodular functions, while maintaining polynomial-time complexity. Extensive simulations demonstrate the algorithm's effectiveness in accurately estimating and tracking targets over extended periods. Furthermore, numerical results confirm that the algorithm's performance is close to that of the optimal assignment, highlighting its robustness and practical applicability.