A communication-bandwidth-aware hybrid estimation framework for multi-robot cooperative localization

Abstract

This paper presents hybrid Minimum Mean Squared Error-based estimators for wireless sensor networks with time-varying communication-bandwidth constraints, focusing on the particular application of multi-robot Cooperative Localization. When sensor nodes (e.g., robots) communicate only a quantized version of their analog measurements to the team, our proposed hybrid filters enable robots to process all available information, i.e., local analog measurements (recorded by its own sensors) as well as remote quantized measurements (collected and communicated by other sensors). Moreover, these filters are resource-aware and can utilize additional bandwidth, whenever available, to maximize estimation accuracy. Specifically, in this paper, we present two filters, the Hybrid Batch-Quantized Kalman filter (H-BQKF) and the Hybrid Iteratively-Quantized Kalman filter (H-IQKF), that can process local analog measurements along with remote measurements quantized to any number of bits. We test our proposed filters in simulations and experimentally, and demonstrate that they achieve performance comparable to the standard Kalman filter.

Active Classification for POMDPs: A Kalman-Like State Estimator

“…the Extended and Unscented KF [21], the Extended Kalman smoother and the Unscented Kalman smoother [25], are suitable for general, nonlinear, (non)-Gaussian systems, and they usually adopt a Gaussian approximation for the state distribution, while their performance depends significantly on either some kind of linearization or the careful selection of samples points. To reduce communication costs in sensor network (SN) applications, the problem of state estimation using quantized observations with/without availability of analog measurements has been addressed [26]- [28]. For example, the proposed Sign-of-Innovation KF [26] and its extensions (see [27], [28] and references therein) are based on quantized versions of the measurement innovation and/or real measurements for both Gaussian linear and non-linear dynamical systems.…”

Distributed Range-Based Relative Localization of Robot Swarms

“…For example [14] extended the EKF approach described in [13] to consider weaker forms of sensors, including distance-only sensing, and [15] described how to reduce the amount of state and communication required. The computational complexity of the EKF was further reduced in [16], and a communication-bandwidth aware solution was described in [17]. Similarly novel techniques to reduce the computational costs of the MCL approach have been proposed, for example [18] described a clustering technique to minimize the amount of state and communication required.…”

Hybrid maximum a posteriori estimation under communication constraints

“…In [12,13], we introduced a hybrid estimation framework for filtering that enables each sensor to obtain MMSE state estimates (under Gaussian assumption) by processing local analog and remote quantized measurements. In this work, we focus on MAP estimation since for the nonlinear process/measurement models in real-world systems, the MAP estimator acts as a smoother and mitigates linearization errors, hence improving estimation accuracy.…”

“…To overcome this drawback, we have introduced a hybrid estimation framework in [12,13], for Minimum Mean Squared Error (MMSE) estimation (filtering), that enables each sensor to incorporate its locally-available, analog observations in the estimation process. Specifically, each sensor maintains two local estimators (see Fig.…”