Rigid Body Localization (RBL) Overview

• Rigid Body Localization is the process of estimating a rigid body’s full 6D pose and structure using fixed geometric constraints.
• It leverages measurements from anchors and onboard sensors with methods like least squares, SDP, and Gaussian belief propagation.
• RBL enhances tracking accuracy in applications such as V2X, robotics, and SLAM compared to conventional point-based localization.

Rigid Body Localization (RBL) is the estimation of a rigid body’s pose—translation and orientation—and, depending on the formulation, its geometric structure and motion, from measurements acquired by anchors, cooperating bodies, or onboard sensing/communication systems. In contrast to conventional point-target localization, RBL models a target as a multi-point object whose constituent points preserve fixed relative geometry under rigid transformations. In contemporary wireless and V2X literature, this shift is motivated by the need to recover not only position and velocity, but also 6D pose, 3D structure, and, in tracking settings, translational and angular velocities (Führling et al., 1 Sep 2025). Early anchor-based formulations treated the problem as joint position and orientation estimation from range-only measurements with known body topology, using rotation matrices and translation vectors constrained by the rigid-body geometry (Chepuri et al., 2013).

1. Conceptual definition and problem scope

RBL models a body not as a single point but as a set of body-fixed landmarks. If the conformation matrix is denoted by $C = [c_1,\ldots,c_K] \in \mathbb{R}^{3\times K}$, then after a rigid transformation the world-frame coordinates satisfy

$S = R C + t\,1_{1\times K},$

with $R \in SO(3)$ and $t \in \mathbb{R}^3$. The defining constraint is invariance of intra-body Euclidean distances: the relative geometry of the body does not change under the pose transformation (Führling et al., 1 Sep 2025).

This formulation separates RBL from classical wireless localization, which estimates a single point $p \in \mathbb{R}^3$ from measurements $z_k = h_k(p)+\text{noise}$ and therefore ignores extent, orientation, and intra-target structure. In RBL, fixed geometry acts as a structural prior: known or partially known conformation constraints reduce ambiguity, lower the number of required measurements, and enable orientation estimation (Führling et al., 2023).

The literature spans several regimes. In anchor-based RBL, sensors mounted on a body measure ranges or other observables to anchors with known positions. In anchorless or egoistic RBL, one rigid body estimates the relative pose of another without external infrastructure and, in some formulations, without prior knowledge of the target conformation (Führling et al., 20 Jan 2025). In distributed multi-agent settings, the objective becomes estimation of each agent’s rigid transformation with respect to a common but unknown reference frame, recoverable only up to a common constant bias when only relative measurements are available (Tran et al., 2019).

A common misconception is that RBL is merely point localization with extra post-processing. The literature does not support that view. RBL changes the estimation problem itself: the unknowns are body pose and sometimes shape or motion states, and the measurements are coupled through rigidity constraints. Another misconception is that anchorless operation implies absolute localization. In fact, relative-measurement formulations identify pose only up to gauge freedom unless anchors or equivalent constraints are introduced (Tran et al., 2019).

2. Geometric and measurement foundations

The canonical kinematic model associates each body-fixed point $p_i$ with a world-frame point

$x_i(t) = R(t)\,p_i + t(t),$

and, in tracking problems, uses

$$\dot{x}_i(t) = v(t) + \omega(t) \times R(t)p_i,\qquad \dot{R}(t) = R(t)[\omega(t)]_\times.$$

This expresses how translational velocity $v(t)$ and angular velocity $S = R C + t\,1_{1\times K},$0 jointly determine the motion of all body points (Führling et al., 1 Sep 2025).

Measurement models depend on modality. In V2X and ISAC settings, standard primitives include range or time-of-flight,

$S = R C + t\,1_{1\times K},$1

angles such as AoA or AoD,

$S = R C + t\,1_{1\times K},$2

and Doppler,

$S = R C + t\,1_{1\times K},$3

where $S = R C + t\,1_{1\times K},$4 is the line-of-sight direction between sensor $S = R C + t\,1_{1\times K},$5 and body point $S = R C + t\,1_{1\times K},$6 (Führling et al., 1 Sep 2025). Related surveys also list ToA, TDoA, RTT, RSS, phase, AoA/DoA, and Doppler as 6G-relevant modalities, with quaternions or Lie-algebra parameterizations used when optimization or filtering is performed directly on orientation variables (Führling et al., 2023).

Range-only formulations remain foundational. Early anchor-based work derived linearized squared-range models and expressed the rigid-body pose as an affine function of a matrix on the Stiefel manifold, leading to least-squares and constrained least-squares estimators under orthonormality constraints on the rotation (Chepuri et al., 2013). Later Gaussian-belief-propagation formulations preserved the range-only assumption but recast the problem into linear Gaussian systems, sometimes under small-angle approximations, to estimate 3D rotation and 3D translation with low complexity (Vizitiv et al., 2024).

For extended targets in radar or ISAC, a widely used narrowband model is

$S = R C + t\,1_{1\times K},$7

with scattering centers tied to body points through the rigid transform. This makes RBL particularly natural for vehicles, which exhibit multiple reflectors whose relative geometry is approximately fixed (Führling et al., 1 Sep 2025).

3. Estimation, optimization, and tracking methods

A generic estimation objective in RBL is

$S = R C + t\,1_{1\times K},$8

possibly with $S = R C + t\,1_{1\times K},$9 and additional rigidity constraints. Maximum-likelihood and MAP variants weight residuals by inverse noise covariance and can incorporate motion smoothness, clock biases, or partial semantic geometry (Führling et al., 1 Sep 2025).

Several algorithmic families recur across the literature. Algebraic methods include multidimensional scaling (MDS) and super-MDS (SMDS), which reconstruct coordinates from distance or distance-plus-angle information and then align them to the known body conformation via rigid transforms. Optimization methods include least squares, semidefinite programming, projections onto convex sets, fractional programming, and matrix completion, especially when Euclidean distance matrix structure can be exploited to reconstruct missing distances under NLOS or occlusion (Führling et al., 1 Sep 2025).

Message-passing methods form a distinct strand. Gaussian belief propagation (GaBP) has been used to estimate absolute sensor positions from range-only data and then recover the rigid-body transformation, as well as to directly infer rotation angles and translation via linearized transformation models. The reported motivation is reduced complexity relative to large-matrix inversions in SDP or centralized least-squares pipelines, together with robustness under noisy or incomplete measurements (Vizitiv et al., 2024). This line extends to moving rigid bodies, where bilinear GaBP combines range and Doppler to estimate not only pose but also translational and angular velocities, followed by interference-cancellation refinements to improve angle estimation (Führling et al., 2024). A later extension replaces the first-order small-angle model with a quadratic angle approximation so that range-based GaBP remains accurate even when the prior orientation is poor or the angular deviation is relatively large (Führling et al., 4 Feb 2026).

Tracking formulations augment pose with velocity states. The surveyed literature explicitly mentions frame-to-frame updates, EKF, UKF, particle filters, and smoothing over time windows, with state vectors such as $R \in SO(3)$0 or their angle-parameterized analogues (Führling et al., 2023). In V2X examples, frame-to-frame rigid-body tracking supports platooning and collision detection by combining per-frame measurements with motion models (Führling et al., 1 Sep 2025).

An important practical issue is data association. Extended targets generate multiple measurements, one per scatterer or body point, and associating measurements to the correct landmark may be solved jointly with pose estimation or through gating and probabilistic association. When full conformation is unavailable, semantic RBL replaces exact geometry with directional or category-level constraints, such as a unit-norm direction vector transformed by $R \in SO(3)$1, to disambiguate shapes and support privacy-aware operation (Führling et al., 1 Sep 2025).

4. Observability, bounds, and performance analysis

Theoretical analysis in RBL is dominated by Fisher-information and Cramér–Rao frameworks. For a parameter vector $R \in SO(3)$2 containing translation, rotation parameters, and possibly velocity or clock terms, the Fisher Information Matrix is written as

$R \in SO(3)$3

with the CRLB

$R \in SO(3)$4

This structure is used in V2X-oriented surveys both as a benchmark for estimator design and as a guide to array, bandwidth, and measurement-type selection (Führling et al., 1 Sep 2025).

A more explicit information-centric construction was developed for rigid bodies by summing measurement-wise rank-one contributions and distinguishing translation and rotation information blocks. That formulation provides closed-form expressions for unconstrained translation and rotation bounds and also a constrained bound for $R \in SO(3)$5 that incorporates orthonormality via a tangent-space projection (Führling et al., 30 Jul 2025). The same analysis shows that angle-type measurements strengthen rotation information much more directly than ranges, whereas translation information scales more strongly with baseline geometry and the number of measurements.

The recurring identifiability conclusions are consistent across papers. Non-collinear or non-coplanar body landmarks and anchors improve observability. Ranges alone are often insufficient for reliable orientation unless enough rigidly coupled multi-point structure is present. Single-anchor angular information leaves ambiguity along rays. Symmetric bodies, collinear anchors, near-parallel rays, and poor angular diversity induce rank-deficient FIMs or numerically weak conditioning (Führling et al., 2023).

The literature also emphasizes incomplete-information regimes. Matrix completion over Euclidean distance matrices can materially improve translation performance when LOS paths are blocked, but the resulting estimators still remain above the corresponding CRLB, indicating residual suboptimality under practical impairments (Führling et al., 1 Sep 2025). In AGV positioning under harsh environments, EDM-completion-based RBL was developed specifically to handle missing TOF measurements by constructing upper and lower bounds for the missing distances from anchor geometry, tag conformation, and measurement statistics, followed by coarse localization, refinement, and rigid-pose recovery (An et al., 2022).

A further misconception is that more measurements automatically ensure full pose observability. The papers instead indicate that measurement diversity and geometry matter at least as much as count. This suggests that RBL system design is fundamentally a joint sensing-and-geometry problem rather than merely a high-redundancy estimation problem.

5. Variants and adjacent formulations

Several important variants extend the basic anchor-based rigid-body model. One is soft-connected wireless RBL, in which multiple rigid objects are modeled jointly: each object obeys rigid intra-body constraints, while inter-object relations are represented by soft constraints such as distance bands, alignment tolerances, or linkage geometry. This generalization is motivated by platooning, swarming, tractor–trailer systems, coordinated industrial assets, and time-coupled cooperative tracking (Führling et al., 2023).

Another variant is egoistic or anchorless RBL. In these formulations, one rigid body estimates the relative pose of another from cross-body distances alone, without infrastructure anchors and, in the strongest version, without knowledge of the target shape. MDS-based methods use double-centering and matrix completion ideas to estimate translation and orientation between bodies of different sizes and shapes (Führling et al., 20 Jan 2025). Robust egoistic extensions explicitly address incomplete observations and propose alternative translation estimators whose complete-information performance is slightly weaker but whose missing-data robustness is stronger (Führling et al., 17 Jan 2025).

Distributed frame-localization formulations occupy a related but distinct position. In the multi-agent SE(3) setting, each agent maintains a body-fixed frame and estimates its transformation with respect to a common but unknown global frame using relative transformations and neighbor communication. Extrinsic estimators evolve auxiliary matrices in Euclidean space and then project back to $R \in SO(3)$6, yielding almost-global convergence modulo a common rigid-body bias and, for undirected connected graphs, finite-time convergence (Tran et al., 2019). This is RBL in the sense of localizing rigid frames rather than localizing a single rigid extended target against anchors.

RBL also appears outside wireless positioning. In RGB-D SLAM with large dynamic rigid objects, joint segmentation, tracking, and reconstruction can be cast as estimation of two rigid transformations: camera motion relative to the static background and object motion relative to the camera. The optimization couples dense photometric and depth alignment with spatial and temporal segmentation regularization, thereby treating the dynamic region as one rigid body rather than as an outlier class (Long et al., 2020). A related SLAM formulation exploits rigid-body pose change directly on tracked feature points, improving robot localization and mapping in dynamic scenes without explicitly estimating object geometry (Henein et al., 2018).

These variants show that RBL is not restricted to a single sensing modality or to a single meaning of “localization.” Depending on the domain, the unknown may be an extended target pose, a relative inter-vehicle pose, a multi-object soft-constrained ensemble, or a dynamic object transformation embedded in a SLAM graph.

6. Applications, standardization, and open research directions

The most emphasized application domain is V2X perception for B5G and 6G. The cited survey literature identifies cooperative perception, platooning, collision avoidance, intersection safety, automatic valet parking, lane merging and maneuvering, and protection of vulnerable road users as central use cases. The key argument is that orientation and shape support safer and more predictive decision-making than point estimates alone, especially when anchorless sidelink-based relative pose estimation is required in sparse-infrastructure environments (Führling et al., 1 Sep 2025).

Industrial and robotic applications are similarly prominent. Smart factories, mobile robots, and digital-twin settings benefit from rigid-body tracking of components or platforms with high precision, while AGV systems benefit from multi-tag rigid-body formulations that maintain localization availability when some anchor–tag paths are blocked (An et al., 2022). In SLAM, large dynamic rigid objects can be exploited rather than suppressed, improving localization and map consistency under substantial occlusion (Long et al., 2020).

Within standardization discourse, RBL is positioned as a prospective 6G capability rather than a fully standardized 5G feature. The literature states that 5G and 5G-A include network-based, sidelink, and hybrid positioning enhancements, including Release 18 sidelink positioning and AI/ML additions, but that a general RBL framework is not yet standardized. At the same time, 3GPP RP-251881 and TR 22.870 are identified as relevant 6G study items because their use cases—environment object reconstruction, smart transportation, cooperating mobile robots, and real-time digital twins—are directly benefited by rigid-body perception (Führling et al., 1 Sep 2025).

Open problems are correspondingly cross-layer. The surveyed directions include converting CRLB insights into optimal sensor and anchor placement; improving matrix completion and joint estimation under NLOS, clutter, and unsynchronized networks; developing scalable low-latency message-passing solvers with convergence guarantees; formalizing egoistic and semantic RBL without prior target conformation; and co-designing ISAC waveforms, arrays, and sidelink resource allocation for extended-target sensing (Führling et al., 1 Sep 2025). Soft-connected many-body localization, RIS-aided observability design, and learning-enhanced measurement models are also identified as active directions in the broader 6G RBL agenda (Führling et al., 2023).

A final limitation remains fundamental: the rigid assumption can be violated. Trailer flex, deformations, articulation, or model mismatch break exact rigidity and necessitate outlier-robust estimators, adaptive models, or soft-constraint generalizations. This suggests that RBL should be viewed not as a universally valid object model, but as a high-fidelity approximation whose utility depends on how well fixed-geometry constraints match the physical target and sensing regime.