CoWear: Towards Collaborative Wearable Inertial Localization with Multiple Smart Devices

Zhuoyuan She; Zhanpeng Zhang; Yapeng Chen; Jiale Wang; Xin Xie; Ming Xia

cs.ROOct 14, 2026

CoWear: Towards Collaborative Wearable Inertial Localization with Multiple Smart Devices

Authors: Zhuoyuan She, Zhanpeng Zhang, Yapeng Chen, Jiale Wang, Xin Xie, Ming Xia

Organizations: School of Electronic and Information Engineering, Beihang University, Beijing, China. · School of Automation Science and Electrical Engineering, Beihang University, Beijing, China. · School of Space and Earth Sciences, Beihang University, Beijing, China. · Jiangxi Research Institute, Beihang University, Nanchang, China.

Abstract

Consumer smart devices observe the same human motion from different body locations, yet existing inertial-localization benchmarks rarely provide synchronized heterogeneous wearables with device-specific motion ground truth. We present CoWear, a dataset and reference framework for collaborative inertial localization using a smartphone, smartwatch, and smart glasses. The dataset contains over 30 participants, over 300 walking sessions, and more than 30 hours and 100 km of aggregate device trajectories, with each device individually tracked by motion capture. To enable collaboration across distinct body motions, we train role-specific estimators to predict the displacement of a common reference device over synchronized temporal windows. Their predictions are transformed to the world frame using gyroscope integration and fused according to learned uncertainty. Experiments using phone, glasses, and watch as reference targets show that multi-device fusion consistently improves both displacement and trajectory rollout accuracy over individual-device estimation. CoWear provides a dataset and reproducible baseline for studying collaborative inertial localization across heterogeneous consumer wearables.

Figures & tables

Fig. 1: Representative device–motion mismatch in RIDI. (a) Ground-truth and NIO trajectories. (b) Motion and device headings. (c) Heading difference. The purple boxes highlight a short interval where device motion deviates sharply from the underlying locomotion.

Fig. 2: Overview of the CoWear dataset construction pipeline. Three heterogeneous wearables are recorded simultaneously and independently tracked by motion capture. Data from multiple participants are temporally aligned, reorganized, and curated into synchronized analysis-ready sessions. Dataset scale is reported using aggregate device trajectories.

Dataset	IMU setup	Duration	Distance	Reference source	Supervision target	Joint streams
RIDI [ 4 ]	1 phone, 4 placements	2.5 h	–	Tango phone	Body trajectory	$\times$
OxIOD [ 12 , 20 ]	1 phone, 4 placements	14.7 h	42.5 km	MoCap / Tango	Phone / body	$\times$
RoNIN [ 2 ]	1 free-carried phone	42.7 h	–	Tracking phone	Body trajectory	$\times$
TLIO [ 3 ]	Headset IMU–camera	–	–	Headset VIO	Headset 6-DoF	$\times$
DIP-IMU [ 17 ]	17 body IMUs	–	–	Optical MoCap	Full-body pose	$\checkmark$ Homogeneous
HelmetPoser [ 19 ]	Helmet-mounted IMUs	–	–	VICON MoCap	Head 6-DoF	$\checkmark$ Homogeneous

TABLE I: Comparison with representative inertial-motion datasets.

Fig. 3: Collaborative localization framework. Each wearable independently estimates the same target displacement and its uncertainty before fusion.

Fig. 4: Representative benchmark trajectories for watch, phone, and glasses sequences. Dashed lines denote ground truth and solid lines denote predictions; per-session ATE is reported in each panel.

	Single Device			Late Fusion
Method	Phone	Glasses	Watch	Fusion
PDR	3.484 / 3.949	2.991 / 3.295	7.011 / 9.486	3.653 / 4.211
RIDI [ 4 ]	1.570 / 1.825	1.710 / 2.016	1.840 / 2.236	1.329 / 1.532
RoNIN [ 2 ]	1.720 / 2.361	2.997 / 3.234	2.137 / 2.668	1.837 / 2.096
TLIO [ 3 ]	1.256 / 1.516	1.543 / 1.647	1.553 / 1.968	1.168 / 1.340
CoWear LSTM	1.389 / 1.686	1.868 / 1.983	1.689 / 1.911	1.313 / 1.441

TABLE II: Trajectory ATE on the test sessions (m, median / mean; lower is better).

Fig. 5: Representative common-target trajectories from individual wearable observations and information fusion. Fusion yields lower per-session ATE than each individual estimate in both examples.

Target	Observation	Disp. RMSE	ATE Med.	ATE Mean
Phone	Phone	0.196	1.440	1.708
	Glasses	0.209	1.934	2.050
	Watch	0.210	1.671	2.004
	Information fusion	0.131	1.173	1.335
Glasses	Phone	0.186	1.408	1.662
	Glasses	0.195	1.868	1.983

TABLE III: Common-target localization across wearable roles (m, lower is better).

Fusion strategy
	Phone ∗	Glasses ∗	Watch ∗
Uniform	1.359 / 1.429	1.294 / 1.390	1.304 / 1.433
Fixed weight	1.308 / 1.411	1.195 / 1.376	1.252 / 1.424
Information	1.173 / 1.335	1.162 / 1.326	1.164 / 1.368
Watch–phone relation
Same hand ( $27$ )	1.218 / 1.410	1.389 / 1.467	1.340 / 1.558

TABLE IV: Collaborative localization analysis (ATE in m, median / mean; lower is better).

Cooperative Localization

Cooperative localization lets robots and other agents estimate their positions by combining onboard motion sensing with shared relative measurements.

See all 42 papers

Towards Real-World Wearable Motion Reconstruction
Jul 8, 2026Andrea Boscolo Camiletto, Rishabh Dabral, Eduardo Alvarado +33D Reconstruction Human Motion Generation
Max Planck Institute for Informatics, Saarland Informatics Campus, Germany · Saarbrücken Research Center for Visual Computing, Interaction and AI, Germany · Google, Switzerland
MARIO: Motion-Augmented Real-Time Multi-Sensor Inertial Odometry
Jun 2, 2026Yiquan Li, Taeyoung Yeon, Chenfeng Gao +3Markerless Motion Capture Inertial Odometry
Northwestern University · University of Chicago
Reliability-Gated Fusion of Consumer Head and Foot IMUs for Lower-Body 3D Pose
Sep 28, 2026Zhilin Guo, Boqiao Zhang, Oszkár Urbán +8Markerless Motion Capture 3D Human Pose Estimation
University of Cambridge · Chalmers University of Technology

Jul 8, 2026J/K move · Enter open · S save

Towards Real-World Wearable Motion Reconstruction

Andrea Boscolo Camiletto, Rishabh Dabral, Eduardo Alvarado, Thabo Beeler+2

Max Planck Institute for Informatics, Saarland Informatics Campus, Germany · Saarbrücken Research Center for Visual Computing, Interaction and AI, Germany · Google, Switzerland

3D Reconstruction Human Motion Generation Wearable Sensing Multimodal Sensor Fusion 3D Human Motion Reconstruction

Open paper AskPDF arXiv

The modern-day surge in popularity of wearable devices poses a fundamentally unique motion capture problem: reconstructing full-body movement from any set of sensing hardware worn at a given moment. Yet, most research efforts assume fixed sensor configurations (e.g. IMU suits or HMD-centric rigs) and cannot generalize across them. In contrast, we argue that motion capture should prioritize unobtrusive and lightweight devices such as smartphones, smartwatches, smart glasses, and smart insoles, and study the interplay between them. To this end, we make three contributions. First, we present a large-scale multi-modal dataset synchronizing these consumer-grade sensors with ground-truth 3D motion, spanning 50 diverse activities including everyday tasks, sports, and social interactions. Second, we propose WHIP, a baseline generative model that reconstructs motion from arbitrary subsets of available sensors, robustly handling missing modalities and producing physically plausible motions. Third, we conduct a systematic study of sensor complementarity, quantifying how different modalities complement one another. Code and dataset are available at https://vcai.mpi-inf.mpg.de/projects/WHIP/

Abstract

Cooperative Localization

Explore similar work

Towards Real-World Wearable Motion Reconstruction

MARIO: Motion-Augmented Real-Time Multi-Sensor Inertial Odometry

Reliability-Gated Fusion of Consumer Head and Foot IMUs for Lower-Body 3D Pose