UniSaT: Unified-Objective Belief Model and Planner to Search for and Track Multiple Objects

TL;DR

This paper introduces UniSaT (Unified Search and Track), a novel unified-objective formulation for the search and track problem based on Random Finite Sets (RFS), and demonstrates both qualitative results and quantitative improvements over a multi-objective method.

Abstract

Path planning for autonomous search and tracking of multiple objects is a critical problem in applications such as reconnaissance, surveillance, and data gathering. Due to the inherent competing objectives of searching for new objects while maintaining tracks for found objects, most current approaches rely on multi-objective planning methods, leaving it up to the user to tune parameters to balance between the two objectives, usually based on heuristics or trial and error. In this paper, we introduce UniSaT (Unified Search and Track), a novel unified-objective formulation for the search and track problem based on Random Finite Sets (RFS). Our approach models unknown and known objects using a combined generalized labeled multi-Bernoulli (GLMB) filter. For unseen objects, UniSaT leverages both cardinality and spatial prior distributions, allowing it to operate without prior knowledge of the exact number of objects in the search space. The planner maximizes the mutual information of this unified belief model, creating balanced search and tracking behaviors. We demonstrate our work in a simulated environment, presenting both qualitative results and quantitative improvements over a multi-objective method.

Figures (5)

• Figure 1: An example scenario of a UAV searching for and tracking objects in a space. The objects are represented by red dots. The UAV is tracking three objects, with two of them currently in the field of view. The Gaussian priors are represented by blue ellipses, and each cluster has an associated cardinality distribution, with the one on the right having a higher cardinality distribution. Using our unified objective function, the UAV planner chooses a path that maximises both tracking and search performance, in this case leaving the tracked targets and searching the area on the right.
• Figure 2: An overview of the UniSaT pipeline. The first step is the belief initialization from the population count, shown on the left. The algorithm then enters the execution loop: observe, update belief, plan, and execute plan. The measurements are represented by purple crosses and the tracks are represented by red dots. With our unified approach, the highest reward plan, highlighted in red, visits the blue density to try to find new objects before visiting one of the tracks to update its state estimation.
• Figure 3: A randomly sampled environment from the five characteristic scenarios and a random environment setup. The true target states are shown in red and the prior SMC particles are plotted in distinct colors. Below each environment plot is the true environment cardinality distribution, which is what we sample the ground truth from, and the prior cardinality, which is used for belief initialization.
• Figure 4: An example run of UniSaT at different time steps in our simulation environment. The range of detection is shown in dotted lines around the UAV, where the probability of detection drops off between black to grey circles. The targets are shown as red dots and the EKF track covariance can be seen around the targets being tracked. SMC belief particles are visualized with each hypothesis having a different color.
• Figure 5: The mean OSPA$^2$ results of UniSaT and the baseline over 100 MC runs on the environment with random configurations, plotted with the $95\%$ confidence interval.