Distributed Kalman--Consensus Filtering with Adaptive Uncertainty Weighting for Multi-Object Tracking in Mobile Robot Networks

2026-03-11

AI Summary

This work addresses estimation inconsistency, trajectory duplication, and ghosting artifacts in multi-object tracking within mobile robot networks operating under partial observability and heterogeneous localization uncertainties, which are exacerbated by frame misalignment. Building upon the MOTLEE framework, the authors introduce a frame alignment mechanism that treats dynamic targets as transient landmarks and propose an uncertainty-aware adaptive consensus weighting strategy that dynamically adjusts the influence of neighboring agents based on their estimation covariances. Locally, each agent employs a Kalman filter with a constant velocity model and global nearest neighbor data association; during fusion, a weighted distributed Kalman consensus filter suppresses unreliable information. Simulations demonstrate that the proposed approach significantly enhances tracking consistency and accuracy under localization drift, improving the MOTA metric by 0.09 for affected agents.

Technology Category

• Intelligent Robots: State Estimation
• Computer Vision: Motion & Tracking
• Multiagent Systems: Multiagent Systems under Uncertainty

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Tracking, profiling, and countermeasures against them

Abstract

This paper presents an implementation and evaluation of a Distributed Kalman--Consensus Filter (DKCF) for Multi-Object Tracking (MOT) in mobile robot networks operating under partial observability and heterogeneous localization uncertainty. A key challenge in such systems is the fusion of information from agents with differing localization quality, where frame misalignment can lead to inconsistent estimates, track duplication, and ghost tracks. To address this issue, we build upon the MOTLEE framework and retain its frame-alignment methodology, which uses consistently tracked dynamic objects as transient landmarks to improve relative pose estimates between robots. On top of this framework, we propose an uncertainty-aware adaptive consensus weighting mechanism that dynamically adjusts the influence of neighbor information based on the covariance of the transmitted estimates, thereby reducing the impact of unreliable data during distributed fusion. Local tracking is performed using a Kalman Filter (KF) with a Constant Velocity Model (CVM) and Global Nearest Neighbor (GNN) data association. simulation results demonstrate that adaptive weighting effectively protects local estimates from inconsistent data, yielding a MOTA improvement of 0.09 for agents suffering from localization drift, although system performance remains constrained by communication latency.

Problem

• Multi-Object Tracking
• Localization Uncertainty
• Frame Misalignment
• Distributed Fusion
• Mobile Robot Networks

Innovation

• Distributed Kalman--Consensus Filter
• Adaptive Uncertainty Weighting
• Multi-Object Tracking
• Frame Alignment
• Mobile Robot Networks