MG-Nav: Unifying Navigation Frameworks

• MG-Nav is a collection of navigation frameworks that use magnetic anomaly, sensor fusion, and astrophysical signals to enable autonomous localization in GPS-denied settings.
• It integrates methods like magneto-inductive localization and entropy-guided planning to reduce uncertainty and improve 3D pose estimation across various environments.
• Experimental validations demonstrate significant error reductions—up to 30% lower uncertainty—across terrestrial and deep-space applications using practical sensor fusion and dynamic programming techniques.

MG-Nav refers to several distinct navigation frameworks unified by the acronym "MG-Nav" in the literature, each designed for information-driven, uncertainty-aware, or memory-guided autonomous navigation under different sensing modalities and environments. This entry surveys the primary lines of published MG-Nav research, with particular focus on magnetic anomaly navigation (MAGNAV), sensor fusion for magneto-inductive localization, information-aware and entropy-based route planning, as well as related developments in multi-modal and memory-based visual navigation.

1. Magneto-Inductive and Magnetic Anomaly-Based Navigation

Magneto-inductive and magnetic anomaly-based navigation frameworks under the MG-Nav moniker exploit spatial variations in naturally occurring or engineered magnetic fields for localization and guidance, particularly where GPS or GNSS signals are unavailable, unreliable, or actively denied.

Magneto-Inductive Navigation

Sensor fusion for magneto-inductive navigation uses quasi-static magnetic dipole fields, with a pre-surveyed or engineered transmitting source, to enable simultaneous 3D position and orientation estimation. The core physical model is governed by

$$B_k = c\,R(\psi)\,\frac{1}{\|r\|^3}\left( \frac{3\,rr^T}{\|r\|^2} - I_3 \right)m_k + e_k$$

where $r$ is the transmitter-receiver displacement, $\psi$ the orientation, $m_k$ the known dipole, $e_k$ Gaussian noise, and $R(\psi)$ the rotation matrix. Observability is further constrained and improved using priors from inertial sensors or geometric knowledge (e.g., coplanarity, attitude constraints).

A maximum a posteriori estimator fuses magneto-inductive and inertial measurements, minimizing nonlinear least squares with regularization by priors on pose and orientation parameters. The Cramér–Rao Bound (CRB) quantifies estimator performance, revealing rapid error growth ($\propto \|r\|^8$) with increasing range, and strong anisotropy favoring estimation in lateral over vertical directions (Wahlström et al., 2019).

To detect environmental distortion (e.g., due to ferrous materials), MG-Nav applies chi-squared and eigenvalue-consistency statistical tests on the residuals and observed channel matrix structure.

Magnetic Anomaly-Based Guidance

MG-Nav exploits spatially textured magnetic anomaly maps, typically surveyed offline, as nonlinear measurement models in the agent's belief update:

$\tilde y_k = h(x_k) + \text{noise}$

with $h(\cdot)$ obtained by look-up from the map. Guidance policies maximize localization observability or minimize expected posterior entropy—formulated via either the determinant of the nonlinear observability Gramian:

$$\mathcal{O}_G = \mathcal{O}_{nl}^T \mathcal{O}_{nl}$$

or the expected information gain (expected entropy reduction) across candidate controls. Receding-horizon dynamic programming or greedy one-step lookahead select headings favoring both progress toward goal and reduced positional uncertainty (Ramos et al., 2022).

Empirical results demonstrate a consistent $20\%$–$30\%$ reduction in position covariance and error compared to shortest-path (information-unaware) guidance, in both simulation and field robotics using low-noise single-axis magnetometers.

2. Entropy-Guided and Multi-Objective Path Planning

A complementary information-driven strategy, also labeled MG-Nav, addresses global path planning via entropy-based assessment of map "informativeness" (Penumarti et al., 2024). The approach constructs an entropy map

$$H(x) = -\sum_{(a,b) \in \text{window}} p_{a,b}(x) \log p_{a,b}(x)$$

where $p_{a,b}(x)$ is the local probability estimate from a sliding-window histogram over normalized field values. Low entropy regions correspond to high spatial-frequency features—interpreted heuristically as high-information or high-observability navigation targets.

The multi-objective planner formulates a potential function over configuration space:

$U(q) = U_{att,G}(q) + U_{att,H}(q)$

with attractive terms driving the agent both toward the global goal $G$ and toward high-information (low-entropy) sites, prioritized by information-theoretic weighting. Planning proceeds by gradient descent on $U(q)$, generating trajectories that minimize global pose uncertainty while ensuring environmental coverage.

In comparison with baseline and local observability-only planners, entropy-driven MG-Nav achieves the lowest and most uniform path-wise entropy, with stable and minimal pose covariance—a robust localization performance even under variation in initial robot heading.

Notably, this entropy-map methodology generalizes beyond magnetics: by swapping in scalar fields such as topography or bathymetry, the framework enables information-aware planning in geophysical domains with analogous spatial structure (Penumarti et al., 2024).

3. Magnetar and Pulsar-Based Space Navigation (MG-Nav)

In the context of deep-space navigation, MG-Nav denotes an overview of autonomous orbital and intersatellite localization using astrophysical time signals from pulsars and magnetars (Luo et al., 2023). The method combines:

• Long-integration absolute navigation using regular pulse time-of-arrival (TOA) measurements from stable millisecond pulsars (e.g., the Crab), exploiting radio ephemerides and barycentric corrections to constrain the spacecraft's position via weighted least-squares:

$$\widehat{\delta\mathbf r} = (H^T \Sigma_t^{-1} H)^{-1} H^T \Sigma_t^{-1} \mathbf{y}$$

where $\mathbf{y}$ are residuals between observed and predicted TOAs, $H$ the known line-of-sight design matrix.

• Inverse triangulation (relative/absolute) using delays from repeated magnetar bursts detected at multiple spacecraft ($j,k$), solved via a similar MLE:

$$\widehat{\mathbf r}_{\rm mag} = (H_{\rm mag}^T\Sigma_{\eta}^{-1} H_{\rm mag})^{-1} H_{\rm mag}^T \Sigma_{\eta}^{-1} \mathbf{d}$$

• Joint estimation couples both modalities for robust orbit and separation estimation. With 16 days of Crab data, absolute navigation down to $\sim20$ km 3D error is demonstrated; with 26 SGR J1935+2154 bursts, inter-craft accuracy of several hundred km is obtained, and further reduced with more bursts.

This method is viable for GNSS-independent space missions requiring fully autonomous state estimation using on-board soft X-ray and gamma-ray detectors.

4. Practical Implementation and Experimental Findings

Across these MG-Nav variants, extensive validation is reported:

• Sensor fusion/Magneto-inductive: Real and simulated 3D positioning tests demonstrate that inertial sensor fusion in MAP estimation nearly halves the median position error versus pure magneto-inductive methods (outdoor errors: $0.40\,\mathrm{m} \rightarrow 0.20\,\mathrm{m}$). The Cramér–Rao bound is approached under well-conditioned scenarios, and well-designed distortion detection reliably flags measurement model breakdown (Wahlström et al., 2019).
• Magnetic entropy-guided planners: Hardware trials using TurtleBot4 mobile robots, total-field magnetometers, and particle-filter localization robustly show $50\%$ reduction in localization covariance and insensitivity to initial conditions for entropy-driven paths (Penumarti et al., 2024).
• Space navigation: Full Fermi/GBM and GECAM datasets yield absolute errors matching theoretical predictions, with code realizable on modest CPUs in minutes for full parameter estimation (Luo et al., 2023).

5. Limitations, Generalization, and Extensions

Principal constraints are:

• Magneto-inductive methods suffer from rapid degradation with range ($\|r\|^8$ dependence in CRB), anisotropy favoring horizontal over vertical accuracy, and distortion sensitivity near ferrous objects (Wahlström et al., 2019).
• Magnetic anomaly navigation relies on the spatial variability of the map; featureless regions lead to poor observability and information gain, explaining trajectory detours toward high-gradient zones (Ramos et al., 2022, Penumarti et al., 2024).
• Space-centric methods depend on the availability and timing stability of astrophysical sources (requiring sufficient integration time for pulsars, and active burst epochs for magnetars) (Luo et al., 2023).

Transferability is strong for entropy and information-driven planners: any scalar gradient field (magnetics, topography, bathymetry) may be substituted with only preprocessing (map normalization, entropy computation) retuned (Penumarti et al., 2024).

Emerging extensions include multi-modal sensor fusion, closed-loop planning in partially observable or dynamic environments, and automatic switching between guidance objectives as uncertainty, information content, or environment dynamics vary.

6. Comparative Summary Table

Below is a high-level comparison of the principal MG-Nav methods:

MG-Nav Variant	Sensor/Modality	Guidance Principle	Main Objective	Typical Accuracy
Magneto-Inductive	Triaxial coils, accelerometer	MAP estimator + CRB	3D pose estimation	$\sim$0.2–0.4 m (short-range)
Magnetic Anomaly (Entropy)	Scalar magnetometer	Entropy-driven gradient	Localization + guidance	20–30% reduced uncertainty
Magnetic Anomaly (Obs.)	Scalar magnetometer	Nonlinear observability	Localization + planning	20–30% reduced uncertainty
Space (Pulsar+Magnetar)	X/gamma-ray detectors	TOA fitting, delay triangulation	Absolute/relative navigation	$\sim$20 km abs. / 100–200 km rel.

• Sensor fusion and CRB for magneto-inductive navigation: (Wahlström et al., 2019)
• Entropy-map/global uncertainty-aware planning: (Penumarti et al., 2024)
• Nonlinear observability and entropy-reduction guidance: (Ramos et al., 2022)
• Pulsar and magnetar navigation for deep-space: (Luo et al., 2023)

These foundational works establish MG-Nav as a meta-class of navigation systems employing information-driven, uncertainty-aware decision making to improve localization and planning under challenging or non-GNSS conditions.