Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

A gravity-aligned radar-leg-inertial pipeline gives legged robots sub-meter vertical odometry on stairs and slopes without LiDAR or cameras.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 21:51 UTC pith:I4ZQ6NXP

load-bearing objection Solid, genuinely novel legged-odometry fusion with a useful open dataset, but the core gravity factor is currently unverifiable and the vertical SOTA claim only covers stationary-start runs. the 3 major comments →

arxiv 2511.13216 v2 pith:I4ZQ6NXP submitted 2025-11-17 cs.RO

GaRLILEO: Gravity-aligned Radar-Leg-Inertial Enhanced Odometry

Chiyun Noh , Sangwoo Jung , Hanjun Kim , Yafei Hu , Laura Herlant , Ayoung Kim This is my paper
classification cs.RO
keywords radar odometrylegged robotgravity estimationcontinuous-time B-splinefactor graph optimizationvertical driftDoppler velocityleg kinematics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tries to establish that a legged robot can estimate its own motion accurately—especially its height on stairs and slopes—using only a low-cost radar, joint encoders, contact sensors, and an IMU, with no LiDAR or camera. The authors argue that the persistent vertical drift in leg-inertial odometry comes from inaccurate roll and pitch, and that this can be fixed by estimating the local gravity vector from velocity rather than from double-integrated IMU acceleration. Their system, GaRLILEO, builds a continuous-time velocity spline from radar Doppler and leg kinematics, then adds a soft S2-constrained gravity factor that keeps the estimated gravity vector at magnitude 9.81 m/s² while optimizing its direction. On twelve real-world indoor and outdoor sequences, they report sub-meter vertical error on most runs—often 10 to 100 times smaller than radar-inertial and leg-kinematics baselines—without any exteroceptive sensing. If this holds, legged robots can navigate visually degraded, feature-sparse environments with cheap proprioceptive-plus-radar sensing while keeping stable vertical pose.

Core claim

Vertical odometry drift in legged robots is mostly a roll-and-pitch observability problem, solvable without LiDAR or cameras by making gravity a continuous-time state. GaRLILEO decouples velocity from the IMU: a B-spline velocity spline is fit to radar Doppler and leg-kinematics velocities, so IMU acceleration observes gravity rather than being double-integrated. A second spline estimates local gravity, and a soft S² factor r_S2 = ḡ + [ω]×g drives its magnitude toward 9.81 m/s² while leaving direction free. With the global gravity reference fixed during stationary startup, the gravity vector gives continuous roll and pitch correction, and pose comes from dead-reckoning the velocity spline. T

What carries the argument

The load-bearing object is a set of three coupled third-order cumulative B-splines: orientation on SO(3), ego-velocity in R³, and local gravity. The named mechanism is the soft S²-constrained gravity factor r_S2 = ḡ + [ω]×g, a first-order derivative term that pulls the gravity spline toward magnitude 9.81 m/s² while letting direction optimize; the authors argue this avoids the brittleness of hard norm constraints in discrete formulations. Velocity is deliberately decoupled from the IMU—radar Doppler and leg kinematics build the velocity spline, so IMU acceleration is not double-integrated—which turns the IMU into a tilt sensor anchored by the radar-leg velocity.

Load-bearing premise

The method needs the robot to stand still at startup so the global gravity vector can be initialized; every evaluation sequence begins stationary, so the claimed sub-meter vertical accuracy has not been shown for a robot that starts walking immediately.

What would settle it

Take the same sensor suite and run GaRLILEO on a trajectory whose first seconds are non-stationary (e.g., the robot is carried or walks immediately after power-on), then compare vertical error to the stationary-initialized run; if vertical error jumps to the level of the no-gravity-factor ablation (several meters), the stationary initialization—not the gravity factor—is the load-bearing element.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Legged robots can maintain vertically accurate odometry on stairs, slopes, and slippery floors using only radar, joint encoders, contact sensors, and an IMU, eliminating the need for LiDAR or cameras in perception-degraded environments.
  • The gravity factor alone drives the largest vertical improvements: removing it raises vertical error on the Quad sequence from about 2.1 m to about 6.9 m, showing that roll and pitch observability is the main lever.
  • Radar and leg kinematics are complementary: in slippery or deformable-floor zones where leg kinematics fails, radar keeps the velocity spline stable, and the fused system outperforms either sensor alone.
  • The continuous-time B-spline formulation allows asynchronous, preintegration-free fusion of 20 Hz radar, 100 Hz IMU, and 150 Hz leg data, mitigating odometry distortion from slips and impacts.
  • Yaw remains weakly observable—horizontal drift still grows on long paths, as the authors acknowledge in the limitations—so the method is strongest for vertical pose, not full 6-DoF localization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the central claim holds, the radar may be replaceable: the soft S² gravity factor needs only a clean continuous velocity signal to anchor tilt, so a high-quality external velocity source (e.g., motion capture or wheel encoders on a different platform) could yield similar vertical gains—a testable experiment.
  • The stationary-startup requirement suggests a natural stress test: running the robot from a moving start, or on a moving platform during initialization, should degrade the global gravity reference and break the sub-meter vertical claim; this boundary is not covered by the paper's evaluations.
  • The same gravity-spline structure could combine with yaw-observing sensors (e.g., a second radar or a magnetometer) to attack the acknowledged horizontal drift, potentially yielding a full 6-DoF proprioceptive odometry that stays vertically and horizontally bounded over long missions.
  • For the SLAM community, the implication is that vertical loop-closure constraints in legged systems may be unnecessary if gravity is estimated well; resources could shift to yaw and horizontal drift instead.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes GaRLILEO, a continuous-time radar-leg-inertial odometry framework for legged robots. Velocity is estimated from a B-spline built from SoC radar Doppler and leg kinematics, decoupled from IMU acceleration; a second B-spline models the local gravity vector, constrained by a soft S2-norm factor and a velocity-aware gravity residual. The global gravity reference is initialized during detected stationary periods and used to align the z-axis before incremental factor-graph optimization. Experiments on 12 self-collected indoor/outdoor sequences with TLS or MoCap ground truth report large reductions in vertical APE relative to several radar-inertial and leg-inertial baselines, with ablations for the gravity factors and the horizontal velocity-bias term. The dataset and code are promised open-source.

Significance. If the derivation is repaired, the reported vertical-accuracy gains are significant: sub-meter APE_z on most sequences, often 5–20× lower than the best baselines, without LiDAR or cameras. The decoupling of velocity from IMU acceleration is a credible mechanism for legged platforms, where contact impacts corrupt accelerometer data, and the soft S2 gravity factor addresses a real numerical brittleness in discrete gravity-magnitude constraints. The TLS/MoCap ground-truth dataset and intended code release would be valuable community assets. The central claim is currently not fully verifiable because the core gravity-factor equation is garbled and references an undeclared magnetometer, and the evaluation is restricted to stationary-start scenarios.

major comments (3)
  1. [§4.4, Eq. (19)] Equation (19), the central gravity factor r_G, is typeset as an unreadable string of glyphs and cannot be checked. The surrounding text references a magnetometer-derived rotation R̃_m and a term β_i^j, but no magnetometer appears in the sensor configuration (Table 1), the state definition (Eq. 10), or the factor graph (Figs. 4 and 6), and β_i^j is never defined. Since this factor is load-bearing for the claimed vertical-accuracy improvement, the paper must provide a clean equation, define every symbol, and either add the magnetometer to the system model or remove it from the formulation.
  2. [§5.1.2, Eqs. (23)–(24)] Global gravity is initialized only under stationary conditions, and the text explicitly states that every sequence in the dataset begins stationary. This means the claimed state-of-the-art vertical accuracy is demonstrated only for stationary startup. If the robot starts while moving, the accelerometer prior is contaminated by linear acceleration and the dynamic term (Eq. 22) is acknowledged to be inaccurate on legged robots; a biased g_G propagates to roll/pitch and vertical drift through Eqs. (33)–(34). This limitation is not listed in §8.1, which mentions only sparse-radar and yaw-drift issues. The claim should be scoped to stationary-start initialization, or experiments with non-stationary startup should be provided.
  3. [§7.5, Table 7] The velocity-bias ablation shows mixed results: on Downstair, Upstair, MoCap-E, and MoCap-H, removing the bias improves APE_xy (e.g., 3.926→3.752, 1.405→0.938, 0.846→0.794, 0.885→0.805), while on other sequences the bias helps. The conclusion is appropriately hedged, but the contributions section presents the velocity bias as a generally beneficial component. At minimum, this mixed evidence should be summarized in the contributions/abstract, and if the bias is intended as a contribution, its failure mode on abrupt slips should be analyzed beyond a qualitative remark.
minor comments (5)
  1. [§3.1, Eq. (1)] The radial-velocity equation is missing the vector dot product/transpose: it should read v_j = -(p_j/‖p_j‖)^T v, matching the later matrix forms in Eqs. (2)–(4).
  2. [Figure 1 caption] Typo: 'GaRLIELEO' should be 'GaRLILEO'.
  3. [Table 6] The ablated variant is labeled 'w/or S2'; this should be 'w/o S2' (without the S2 factor).
  4. [§4.4 and throughout] The notation 'S2' should be typeset as S² consistently, and the factor is described as 'soft S2-constrained' before its equation is introduced; a brief definition of 'soft' would help.
  5. [§5.1.2] The terms 'static initialization' and 'stationary initialization' are used interchangeably; please standardize the terminology.

Circularity Check

0 steps flagged

No significant circularity; vertical-accuracy claim rests on independent radar/leg velocity and IMU measurements, with a scoping limitation on stationary startup.

full rationale

GaRLILEO's derivation chain is not circular. The velocity state used for odometry is constructed from radar Doppler (Eq. 14) and leg forward-kinematics (Eqs. 15-18), and the incremental objective (Eq. 28) omits the accelerometer factor, so the velocity spline is not fit to the vertical pose it later predicts. The global gravity reference is a fixed physical constant (Eq. 10), obtained by a stationary accelerometer prior plus a dynamic term (Eq. 24), not tuned to APE_z. The soft S2 factor (Eq. 20) and post-optimization residual (Eq. 33) enforce consistency between the gravity spline and rotation, both of which are driven by independent IMU gyro/accel and radar/leg measurements. Evaluation uses independent TLS/MoCap ground truth (Sec. 6.4), and baselines are external. Self-citations to GaRLIO and Co-RaL are motivational and provide baselines only; there is no uniqueness theorem or ansatz whose only support is a same-author citation. The one noteworthy assumption - Section 5.1.2 admits 'every sequence included in the dataset of this paper begins stationary, thereby enabling the use of static initialization' - is a domain restriction on the claimed vertical SOTA, not a reduction of the output to an input; it should be weighed as a correctness/generalization limitation (and is omitted from Sec. 8.1), but it does not make the derivation circular.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The central claim rests on standard factor-graph and spline machinery plus two domain assumptions (stationary start; horizontal-only slip). The main free parameters are the unreported optimization weights and the online-estimated biases. No new physical entities are introduced.

free parameters (3)
  • factor graph weights and stationary thresholds
    The optimization in Eq. (28) uses weights w_g, w_S2, w_ω, w_L, w_R, w_b, w_p, w_e, and Eq. (23) uses thresholds τ1, τ2 for stationary detection. Values are not given, and tuning is not described; results may depend on these choices.
  • velocity bias b_v = time-varying 2D state, estimated online
    Eq. (18) introduces a horizontal-only velocity bias to absorb leg-slip; its x-y values are fitted per window to reconcile radar and leg velocity. This is a modeling parameter, not an external constant.
  • accelerometer bias b_a = estimated online
    Accelerometer bias is modeled as Gaussian random walk and estimated in the factor graph; standard for IMU fusion but still a fitted quantity.
axioms (4)
  • domain assumption A stationary interval exists at startup for global gravity initialization (Eq. 23-24).
    Sec 5.1.2 explicitly says all dataset sequences begin stationary; the static initialization is required for the global gravity reference. Non-stationary starts are not evaluated.
  • domain assumption Leg contact frames remain static during contact, up to a horizontal slip compensated by b_v (Eq. 16-18).
    The leg kinematics velocity is derived under no-slip; slip is partially absorbed by the horizontal velocity bias. Vertical contact-frame motion (cushions, dragged floor) is not modeled.
  • domain assumption Radar Doppler targets are predominantly static; outliers are handled by RANSAC/Cauchy loss.
    Eq. (1)-(4) assume stationary scatterers; Sec 7 notes dynamic objects and sparse returns in Quad. The Cauchy loss ρ_r in Eq. (28) mitigates but does not remove the assumption.
  • standard math B-spline theory as presented in Sommer et al. 2020; Schur complement marginalization (Sibley et al. 2010); standard least-squares and factor-graph optimization.
    These are invoked in Sec 3.3, 4.1, 5.2.2 without proof; accepted background.

pith-pipeline@v1.3.0-alltime-deepseek · 34570 in / 15472 out tokens · 147663 ms · 2026-08-03T21:51:06.720372+00:00 · methodology

0 comments
read the original abstract

Deployment of legged robots for navigating challenging terrains (e.g., stairs, slopes, and unstructured environments) has gained increasing preference over wheel-based platforms. In such scenarios, accurate odometry estimation is a preliminary requirement for stable locomotion, localization, and mapping. Traditional proprioceptive approaches, which rely on leg kinematics sensor modalities and inertial sensing, suffer from irrepressible vertical drift caused by frequent contact impacts, foot slippage, and vibrations, particularly affected by inaccurate roll and pitch estimation. Existing methods incorporate exteroceptive sensors such as LiDAR or cameras. Further enhancement has been introduced by leveraging gravity vector estimation to add additional observations on roll and pitch, thereby increasing the accuracy of vertical pose estimation. However, these approaches tend to degrade in feature-sparse or repetitive scenes and are prone to errors from double-integrated IMU acceleration. To address these challenges, we propose GaRLILEO, a novel gravity-aligned continuous-time radar-leg-inertial odometry framework. GaRLILEO decouples velocity from the IMU by building a continuous-time ego-velocity spline from SoC radar Doppler and leg kinematics information, enabling seamless sensor fusion which mitigates odometry distortion. In addition, GaRLILEO can reliably capture accurate gravity vectors leveraging a novel soft S2-constrained gravity factor, improving vertical pose accuracy without relying on LiDAR or cameras. Evaluated on a self-collected real-world dataset with diverse indoor-outdoor trajectories, GaRLILEO demonstrates state-of-the-art accuracy, particularly in vertical odometry estimation on stairs and slopes. We open-source both our dataset and algorithm to foster further research in legged robot odometry and SLAM. https://garlileo.github.io/GaRLILEO

Figures

Figures reproduced from arXiv: 2511.13216 by Ayoung Kim, Chiyun Noh, Hanjun Kim, Laura Herlant, Sangwoo Jung, Yafei Hu.

Figure 1
Figure 1. Figure 1: Overall preview of GaRLILEO. The four subfigures in the upper row present the problematic situations that quadrupedal robots may encounter while performing real-world tasks, while the yellow letters specify the situations and the red words explain the substantial issues generated from them. Two boxes in the left part of the lower row summarize the major contribution and method of the GaRLIELEO, which signi… view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between body-centric and contact-centric leg locomotion of a single leg during the contact state. (a) On body-centric calculation, the end-effector position differs as time passes. Based on the contact information from the contact sensor positioned at every foot, every contact frame should remain static while the contact sensor is on. Therefore, using the forward kinematics, the ego-centric velo… view at source ↗
Figure 3
Figure 3. Figure 3: Soft S 2 -constrained gravity factor rS2 . Prior to optimization, the orange vectors are initialized through gravity spline extrapolation. During optimization, the factor (blue) constrains vectors that would otherwise drift in the R 3 manifold (red), steering them to lie on the S 2 surface (green). kinematics factor rL and the radar factor rR are defined as rL(ti) = (R I B) ⊺ (R¯ I0 G v(ti) + ⌊ω I (ti)⌋×t … view at source ↗
Figure 6
Figure 6. Figure 6: Factor Graph Overview. At each iteration, the number of control points marginalized (grey) equals the number newly added in the preceding window (red); the remaining control points are carried forward as a prior factor for the next solve (blue). The marginalized control points (grey) are then subjected to a post-optimization stage that refines the SO(3) spline. 5.1.4 Z-Axis Alignment with Global Gravity Th… view at source ↗
Figure 7
Figure 7. Figure 7: SNU and RAI sensor system deployment. Both systems include the same TI-mmWave radar, Microstrain IMU, and Boston Dynamics Spot quadrupedal robot, but are attached with different extrinsics [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Environmental examples of the acquired sequences. Diverse environments are included in each sequence to consider various situations that the quadrupedal robot may encounter in a real-world mission. • CorriLoop: A narrow, rectangular indoor corridor traversed twice; long straight segments with repeating doors/walls and a glossy floor create perceptual aliasing, while four sharp 90° turns stress turn handlin… view at source ↗
Figure 9
Figure 9. Figure 9: Examples of ground truth TLS map on SNU sequences. Leveraged for generating ground truth trajectory. et al. 2025a), unsuitable. While MoCap systems can provide high-frequency and high-precision pose estimates, their limited workspace makes them impractical for large￾scale deployments (Doer and Trommer 2021). Some prior works (Tranzatto et al. 2022b) utilize survey-grade maps as ground truth references by p… view at source ↗
Figure 10
Figure 10. Figure 10: Radar–Inertial baseline odometry results on the (a) BridgeLoop and (b) SlopeStair sequences. Red dotted lines in (a, bottom) indicate the starting points of each loop in BridgeLoop sequence, while the faint yellow region in (b, top) denotes the outdoor segment of the SlopeStair sequence. The red zoomed-in views highlight that GaRLILEO and Co-RaL—both integrating SoC radar and leg kinematics—converge more … view at source ↗
Figure 11
Figure 11. Figure 11: Leg-Kinematics baseline odometry results on the (a) CorriLoop and (b) Upstair sequences. Red dotted lines in (a, bottom) indicate the starting points of each loop in CorriLoop sequence. Both red zoomed-in views in (a) and (b) highlight that GaRLILEO integrating SoC radar and leg kinematics converge more closely to the ground-truth final position. The colored numbers on the right-hand side of the bottom pl… view at source ↗
Figure 12
Figure 12. Figure 12: xy plot and z-time plot of MoCap sequences. (a) and (b) are xy and z-time odometry plot of MoCap-E and MoCap-H sequences, respectively. Blue dotted box: zoomed view of slippery zone where leg odometry fails temporarily, especially on the xy plane. The yellow-highlighted area is the exact slippery zone. Comparing the radar-only, leg kinematics-only, and full module versions of GaRLILEO to check the coopera… view at source ↗
Figure 13
Figure 13. Figure 13: vx-time and vz-time Velocity Error (difference from ground truth) in the MoCap-E and MoCap-H sequences. A dotted square on the (a) and (b) includes a slippery zone where the floor moves backwards, while two dotted squares on the (c) and (d) include cushion zones where the leg kinematics velocity presents a high impact on the vertical direction. than other versions of GaRLILEO. Still, the combination of bo… view at source ↗
Figure 14
Figure 14. Figure 14: Effect of local gravity factors on vertical pose estimation. Vertical odometry accuracy is notably enhanced when both local gravity factors are added to the system. This result supports the idea that precise local gravity estimation can provide roll and pitch observations, thereby mitigating vertical drift in odometry. local gravity factors notably improves odometry accuracy in the vertical direction. Spe… view at source ↗
Figure 16
Figure 16. Figure 16: xy plane plot of odometry estimation with and without velocity bias. On Atrium, SlopeStair, and CorriLoop sequences, which are presented in (a), (b), and (c), velocity bias makes odometry more robust as radial velocity information of each radar target point mitigates the error of leg velocity occurring from inaccurate contact and joint encoder measurement. A more detailed zoomed view of each subfigure, sh… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Online Learning of Robust Legged Odometry with Minimal Exteroceptive Supervision

    cs.RO 2026-06 unverdicted novelty 5.0

    Develops an online-learned proprioceptive velocity model supervised by exteroceptive pipelines, fused via InEKF with IMU for resilient quadruped odometry that adapts without explicit calibration.

  2. Four Simple Proprioceptive Estimators for Legged Robots

    cs.RO 2026-05 unverdicted novelty 5.0

    Presents four contact-aided floating-base state estimators for legged robots progressing from invariant EKF to factor-graph updates and fixed-lag smoothers, with open implementations.

Reference graph

Works this paper leans on

7 extracted references · 4 linked inside Pith · cited by 2 Pith papers

  1. [1]

    (2021) Nebula: Quest for robotic autonomy in challenging environments; team costar at the darpa subterranean challenge

    Agha A, Otsu K, Morrell B, Fan DD, Thakker R, Santamaria- Navarro A, Kim SK, Bouman A, Lei X, Edlund J et al. (2021) Nebula: Quest for robotic autonomy in challenging environments; team costar at the darpa subterranean challenge. arXiv preprint arXiv:2103.11470. Arndt C, Sabzevari R and Civera J (2023) Do planar constraints improve camera pose estimation ...

    Pith/arXiv arXiv 2021

  2. [6]

    pp. 607–612. Prepared usingsagej.cls xxiv Journal Title XX(X) Seo DU, Lim H, Lee S and Myung H (2022) Pago-loam: Robust ground-optimized lidar odometry. In:Proceedings of the International Conference on Ubiquitous Robots. IEEE, pp. 1–

    2022

  3. [7]

    In:Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems

    Shan T and Englot B (2018) Lego-loam: Lightweight and ground- optimized lidar odometry and mapping on variable terrain. In:Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems. pp. 4758–4765. Shu F, Xie Y , Rambach J, Pagani A and Stricker D (2021) Visual slam with graph-cut optimized multi-plane reconstruction. In: Proce...

    Pith/arXiv arXiv 2018

  4. [15]

    Qin T, Li P and Shen S (2018) Vins-mono: A robust and versatile monocular visual-inertial state estimator.IEEE Transactions on Robotics34(4): 1004–1020

    Springer. Qin T, Li P and Shen S (2018) Vins-mono: A robust and versatile monocular visual-inertial state estimator.IEEE Transactions on Robotics34(4): 1004–1020. Ramezani M, Khosoussi K, Catt G, Moghadam P, Williams J, Borges P, Pauling F and Kottege N (2022) Wildcat: Online continuous-time 3d lidar-inertial slam.arXiv preprint arXiv:2205.12595. Roston G...

    Pith/arXiv arXiv 2018

  5. [68]

    Chen H, Liu Y and Cheng Y (2023) Drio: Robust radar-inertial odometry in dynamic environments.IEEE Robotics and Automation Letters8(9): 5918–5925. Chen S, Li X, Li S, Zhou Y and Wang S (2024) River: A tightly- coupled radar-inertial velocity estimator based on continuous- time optimization.IEEE Robotics and Automation Letters9(7): 6107–6114. Dh´edin V , L...

    2023

  6. [1710]

    Grupp M (2017) evo: Python package for the evaluation of odome- try and slam.https://github.com/MichaelGrupp/ evo

    Geiger A, Lenz P, Stiller C and Urtasun R (2013) Vision meets robotics: The kitti dataset.International Journal of Robotics Research32(11): 1231–1237. Grupp M (2017) evo: Python package for the evaluation of odome- try and slam.https://github.com/MichaelGrupp/ evo. Harlow K, Jang H, Barfoot TD, Kim A and Heckman C (2024) A new wave in robotics: Survey on ...

    arXiv 2013

  7. [2270]

    In:Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems

    Michalczyk J, Jung R and Weiss S (2022) Tightly-coupled ekf- based radar-inertial odometry. In:Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems. pp. 12336–12343. Nemiroff R, Chen K and Lopez BT (2023) Joint on-manifold gravity and accelerometer intrinsics estimation for inertially aligned mapping. In:Proceedings of th...

    Pith/arXiv arXiv 2022