Nonlinear Estimation of a Faulty Agent in a Multiagent System Using Relative Pose Observations

This study addresses the problem of estimating the states of a rigid-body agent that has lost the ability to sense or estimate its own state, referred to as a faulty agent, using only relative-pose (position and attitude) measurements from vision-based sensors onboard other agents in a multiagent system. A single observer may suffer from poor or intermittent observability due to limited sensor coverage, environmental constraints, or large relative separations. By leveraging multiple supporting agents communicating through a connected information network, the estimation scheme benefits from diverse viewing geometries, prolonged coverage, and redundant measurements, providing a more reliable and accurate basis for recovering the faulty agent’s state. The relative motion dynamics and estimation scheme are formulated on the Lie group SE(3) and its tangent bundle TSE(3), naturally capturing the nonlinear coupled translational and rotational dynamics between multiple rotating and translating reference frames. A square-root unscented Kalman filter on TSE(3) is adapted to operate with time-varying measurement availability governed by sensor field-of-view constraints and a measurement-graph topology, while multiagent fusion is achieved through covariance-weighted averaging and a Karcher mean on SO(3). This formulation enhances estimation fidelity, preserves the underlying geometric structure of the state space, and supports a scalable multiagent framework for fault-tolerant vehicle operations.