Multi-Objective Multi-Agent Planning for Discovering and Tracking Multiple Mobile Objects

TL;DR

The paper tackles online planning for a multi-agent system with limited field-of-view to discover and track an unknown, time-varying number of moving objects under uncertain data associations. It formulates a centralized MPOMDP using random finite sets and the MS-GLMB filter to maintain a tractable information state, and develops differential-entropy-based tracking and occupancy-based discovery value functions. By combining these via a multi-objective framework and leveraging a greedy algorithm, it achieves a submodular optimization with a provable (1-1/e) optimality bound and favorable computational scaling. Experiments on the CRAWDAD taxi dataset with up to 10 UAVs validate improved tracking performance and scalable planning time, demonstrating practical impact for real-time, multi-object MOT with uncertain origins.

Abstract

We consider the online planning problem for a team of agents to discover and track an unknown and time-varying number of moving objects from onboard sensor measurements with uncertain measurement-object origins. Since the onboard sensors have limited field-of-views, the usual planning strategy based solely on either tracking detected objects or discovering unseen objects is inadequate. To address this, we formulate a new information-based multi-objective multi-agent control problem, cast as a partially observable Markov decision process (POMDP). The resulting multi-agent planning problem is exponentially complex due to the unknown data association between objects and multi-sensor measurements; hence, computing an optimal control action is intractable. We prove that the proposed multi-objective value function is a monotone submodular set function, which admits low-cost suboptimal solutions via greedy search with a tight optimality bound. The resulting planning algorithm has a linear complexity in the number of objects and measurements across the sensors, and quadratic in the number of agents. We demonstrate the proposed solution via a series of numerical experiments with a real-world dataset.

Figures (10)

• Figure 1: An unmanned aerial vehicles (UAV) team tracking multiple vehicles with limited FoV sensors and unknown measurements-to-objects associations.
• Figure 2: A schematic of the proposed planning approach.
• Figure 3: CRAWDAD taxi dataset: $10$ taxis travelling over a $1000$ s period, within an area of $6150~\text{m}\times6080$ m. The symbols $\bigcirc/\Box$ indicate the Start/Stop positions of each taxi, the different colours representing different taxis
• Figure 4: Tracking performance of multi-objective planning with $V_{mo}$ versus baseline methods in $100$ Monte Carlo runs as $|\mathcal{N}|$ increases from $2$ to $10$. $V_{mo}^{*}$ represents the optimal performance under ideal conditions where measurement origins are known.
• Figure 5: Average planning time of multi-objective planning with $V_{mo}$ versus baseline methods in $100$ Monte Carlo runs as $|\mathcal{N}|$ increases from $2$ to $10$.

Theorems & Definitions (19)

• Remark 1
• Remark 2
• Definition 1
• Remark 3
• Remark 4
• Proposition 1
• Remark 5
• Remark 6
• Proposition 2
• Definition 2