Inertial odometry (IO) using only Inertial Measurement Units (IMUs) enables camera-free head tracking in head-mounted devices such as VR headsets, AR glasses, and earbuds. Prior IO methods operate in a gravity-aligned world frame, which is not yaw-equivariant. Yaw augmentation improves robustness but does not guarantee equivariance; learned equivariant layers achieve it, but EqNIO's full pipeline costs nearly 3× the FLOPs of ours. While raw body-frame IMU inputs are inherently yaw-invariant, head tilt redistributes both gravity and motion signals across device axes, making the representation tilt-variant. We propose HeadIO, which combines a body-frame heading decomposition with a bidirectional Mamba backbone for head-mounted inertial odometry. Our decomposition achieves tilt invariance by separating gravity-aligned and horizontal components using only the tilt component of device orientation. We evaluate on four head-mounted datasets and find that HeadIO improves RTE-5s by 8.1% (0.260 m → 0.239 m) over world-frame methods with learned equivariant layers on Nymeria, the largest benchmark dataset, at nearly 3× lower FLOPs and 8× fewer parameters. HeadIO achieves state-of-the-art performance on our primary metric, RTE-5s, on all Aria-glasses datasets, which natively provide body-frame semantics for our decomposition method. We further evaluate the IMU-only deployment scenario on Nymeria, where HeadIO remains state-of-the-art. We show that our method relies mainly on tilt accuracy, which can be obtained from the gravity signal; for yaw, a single-coefficient gyroscope correction removes 11.6% of the yaw error and improves RTE-5s by 2.5%, and sparse external yaw corrections (for example, one every 30 s) can improve RTE-5s by 12.0%. Our decomposition is lightweight, composes with any backbone, and is suitable for real-time deployment on head-mounted devices.
Figures & tables
Fig. 1: Invariance properties of the body-frame heading decomposition. (a) Body-frame IMU inputs are heading-invariant: the same walking motion produces identical IMU readings regardless of which direction the user faces. (b) Our heading decomposition additionally achieves tilt invariance: as the head pitches or rolls, the gravity-aligned basis adapts so that the forward, lateral, and vertical components remain consistent across head orientations.
Fig. 2: Body-frame heading decomposition. (A) When the head is upright, the body-frame up axis (+y) aligns with gravity direction eg , and the forward axis fb coincides with eh1 . (B) When the head tilts, fb gains a gravity-aligned component, causing gravity leakage into the forward channel. Our decomposition projects fb onto the plane perpendicular to eg (Eq. 4 ), recovering a tilt-invariant forward direction eh1 .
Dataset
Method
RTE-5s ↓
RTE-1s ↓
ATE ↓
Drift% ↓
Nymeria
TLIO
0.278
0.101
4.093
2.17
EqNIO
0.260
0.093
3.577
1.96
HeadIO
0.239
0.086
3.258
1.79
AEA
TLIO
0.241
0.088
0.811
5.24
EqNIO
0.220
0.081
0.797
5.14
HeadIO
0.213
0.077
0.819
5.81
TABLE I: Main results on the four head-mounted datasets (mean over test sequences; best per metric in bold ). RTE and ATE in meters, drift in percent.
Backbone
Frame
Prep.
Prior
Ground truth
Mahony (IMU-only)
RTE-5s
RTE-1s
ATE
Drift%
RTE-5s
RTE-1s
ATE
Drift%
ResNet
World
–
TLIO
0.278
0.101
4.093
2.17
0.310
0.108
4.514
2.20
VN
EqNIO
0.260
0.093
3.577
1.96
0.297
0.102
4.118
2.05
Body
–
AirIO
0.285
0.103
4.806
2.68
0.314
0.109
5.493
2.88
Ours
Ours
0.260
0.095
3.448
1.84
0.294
0.103
3.987
1.96
Mamba
World
–
TLIO
0.277
0.096
5.159
3.00
0.309
0.104
5.570
3.03
TABLE II: Input representation × backbone at the 3s window on Nymeria under ground-truth and Mahony (IMU-only) orientation ( n=110 , mean over sequences). RTE and ATE in meters, drift in percent; best per column within each backbone in bold .
Fig. 3: Per-sequence CDFs on Nymeria under ground-truth orientation ( n=110 ).
Fig. 4: Validation MSE versus training epoch for the four input representations on both backbones (Nymeria, 3s window).
Fig. 5: Two Nymeria test sequences. (a, c) Estimated trajectories under ground-truth orientation. (b, d) The same sequences under IMU-only orientation: HeadIO without correction (dashed) and with one external yaw fix every 30 s (solid).
Orientation
Method
RTE-5s ↓
RTE-1s ↓
ATE ↓
Drift% ↓
Ground truth
TLIO
0.278
0.101
4.093
2.17
EqNIO
0.260
0.093
3.577
1.96
HeadIO
0.239
0.086
3.258
1.79
Mahony (0.35 / 4.21)
TLIO
0.310
0.108
4.514
2.20
EqNIO
0.297
0.102
4.118
2.05
HeadIO
0.275
0.094
3.914
1.96
TABLE III: Orientation-source study on Nymeria ( n=110 , mean over sequences). Each IMU-only source is listed with its own tilt / yaw error (°). RTE and ATE in meters, drift in percent; best per metric within each source in bold .
Inertial odometry (IO) using only Inertial Measurement Units (IMUs) provides a lightweight solution for human motion tracking in augmented reality (AR) and wearable devices. Recent learning-based IO methods have improved the generalizability of inertial localization through large-scale pretraining on human motion datasets. However, these approaches remain prone to drift and noise because they do not explicitly capture human motion dynamics, especially on daily activity datasets such as Nymeria. In this work, we propose to ground inertial odometry in human kinematics through a learned IMU-inferred pose prior, which promotes physically consistent motion constraints. We integrate this pose prior into existing IO architectures and reduce positional drift by up to 36% on the challenging Nymeria dataset, which is 5x larger than datasets used in prior work. We further improve long-term performance with a sensor-fusion framework that incorporates auxiliary signals from lightweight sensors already available on commercial AR glasses, including magnetometers, barometers, and secondary IMUs. With this fusion strategy, positional drift is reduced by up to 42%, improving robustness and generalization across diverse motion conditions. Together, our results introduce a new paradigm for inertial and lightweight odometry by unifying human motion kinematics with multimodal sensing, setting a new benchmark for accurate and robust camera-less human tracking. Our website is available at https://spice-lab.org/projects/MARIO/.
Neural inertial odometry increasingly uses networks as learned measurements inside filtering pipelines. Such measurements should transform consistently under arbitrary IMU mounting conventions: their mean must transform as a vector, and their covariance must transform congruently as a second-order tensor. We present GINIO, a geometric SO(3)-equivariant interface for neural inertial odometry under arbitrary rotations of the IMU measurement frame. Given calibrated IMU windows, our framework predicts a motion measurement and uncertainty obeying these tensorial laws. To support efficient sensor-frame learning, we introduce Last-Frame Alignment (LFA), a deterministic preprocessing step that is provably equivalent to world-frame training for SO(3)-equivariant predictors. The connected estimator tracks sensor-local states such as IMU bias, separating nuisance estimation from the geometric law enforced by the learned measurement. We instantiate the same interface in filter-connected NIO, AirIO-style recurrent aerial prediction, EqNIO-style full-SO(3) canonicalization, and ResNet-style temporal backbones. On TLIO, GINIO achieves 2.018 m ID/SO(3) ATE while EqNIO degrades to 76.389 m, using 11.6x fewer FLOPs. On NanoBench, our AirIO-style instantiation improves ATE from 5.579 m to 1.430 m without external attitude input, and our ResNet-style instantiation reaches 0.581 m ATE versus 0.645 m for ResNet1D. On Fetch, GINIO empirically reduces unseen physical-remount ATE from 8.15 m to 0.50 m without retraining, demonstrating robustness beyond the exact coordinate-frame guarantee. For uncertainty, spectral covariance reduces covariance-equivariance error by over three orders of magnitude compared with a diagonal head.
Chankyo Kim, Minghan Zhu, Tzu-Yuan Lin +2
University of Michigan Ann Arbor, MI, United States
The training of learned inertial odometry depends on dense, high-precision position ground truth from motion capture, visual-inertial odometry or SLAM, which is costly and hard to acquire at scale. We propose DINS-IO, which learns inertial odometry directly from raw IMU streams without position labels. Our key insight is that the strapdown INS velocity recursion is a strong, fully differentiable consistency prior: the predicted velocity, rotated into the navigation frame, must agree with the integrated specific force up to an unknown initial velocity and a constant accelerometer bias. We cast this constraint as a sliding-window least-squares problem with a globally shared bias, solve it in closed form, and use the solver residual as a self-supervised loss whose gradient flows back to the network through the analytic solution. To supply this per-sample constraint, we design a high-frequency network that emits dense body-frame velocity at the IMU rate. Since the self-supervised network learns consistent motion but its velocity is not yet metrically calibrated, we calibrate it to true metric velocity from a few labeled trajectories by directly supervising the predicted body-frame velocity and adapting only low-rank (LoRA) patches. On standard benchmarks, DINS-IO pretrained self-supervised and fine-tuned with a small fraction of labels matches or surpasses fully supervised baselines.
Hao Qiao, Yan Wang, Jian Kuang +1
School of Geodesy and Geomatics, Wuhan University · GNSS Research Center, Wuhan University · Hubei Luojia Laboratory