Magnetic based In-situ Self 3D Pose Estimation for a Modular Soft Tendon-Driven Continuum Robot via IMU-Fusion

Zheng Cao; Guo Ning Sue; Xiangyun Bu; David Quinn; Junzhe Hu; Carmel Majidi

cs.ROSep 30, 2026

Magnetic based In-situ Self 3D Pose Estimation for a Modular Soft Tendon-Driven Continuum Robot via IMU-Fusion

Authors: Zheng Cao, Guo Ning Sue, Xiangyun Bu, David Quinn, Junzhe Hu, Carmel Majidi

Organizations: Carnegie Mellon University

Abstract

Continuum robots are well suited for gentle manipulation because of their inherent compliance and ability to adapt to complex environments. However, their continuously deformable structure makes accurate configuration estimation challenging, particularly when external vision systems are unavailable or obstructed. In this work, we present an embedded pose sensing framework that combines inertial measurement units (IMUs) and active magnetic fields to estimate the robot configuration without relying on external cameras. The angular measurements from the IMU and magnetic-field references are fused to improve local orientation estimation and reduce accumulated orientation error during operation. This pose sensing scheme achieves an update rate of 16.7~Hz, allowing real-time feedback. The proposed system is experimentally validated through closed-loop control, where the estimated robot configuration is used to maintain the end-effector at a desired position while interacting with an object. These results demonstrate the potential of distributed magnetic--inertial sensing for real-time pose estimation and closed-loop control of continuum robots.

Figures & tables

Fig. 1: Modular tendon-driven continuum robot and sensing principle. a) Pose estimation during 3D curling. Frontal images show the physical robot, while isometric views highlight the reconstructed three-dimensional backbone. Orange denotes our estimate and black denotes motion-capture ground truth. b) Modular platform design and sensing principle. Relative motion between identical segments is modeled using virtual pivots, with relative positions inferred from magnetic fields generated by embedded PCB coils. c) Grasping demonstration with a soft, deformable object (T-shirt).

Fig. 2: Overview of the mechanical and electronic components of one proprioceptive module of the continuum robot. a) Exploded view of a single module. Numbered parts include (1) PCB (2) coil (3) module (4) support standoff (5) M2-20mm screw (6) M2 nut. b) Customized sensing boards in each module and customized coil.

Fig. 3: Sensing principle, data collection setup, and sampled pose distribution. a) Joint bending and magnetic sensing under the fixed pivot assumption of [ 13 ] . The IMU and magnetometer on the upper segment are assumed to move along a hemispherical surface at a fixed radius from a virtual pivot at the center of the compliant joint. b) Data collection setup using two cameras to track a world reference tag and a joint tag attached to the upper segment. The world tag provides a common reference frame, allowing the joint position to be measured when its tag is visible to either camera. Random target orientations guide sampling across the reachable workspace to learn the mapping from magnetic measurements to joint position. c) Sampled pose distribution showing (i) positions across the hemispherical workspace, (ii) a fitted spherical surface with an RMSE of 0.19 mm, supporting the fixed pivot assumption, and (iii, iv) histograms of colatitude and azimuth, respectively. Sampling covers the workspace, although hardware biases and differences between camera measurements may contribute to uneven coverage.

Fig. 4: Pose estimation during a) planar curling, b) obstacle contact, c) free 3D curling, and d) manual manipulation. Panels (i) to (iv) show selected robot configurations and the corresponding estimated backbones. Panel (v) shows instantaneous backbone RMSE against motion capture, with errors at the final frame annotated. The fused estimate (our estimation) is shown in orange, the magnetic estimate as a dashed line, constrained acceleration with quaternions as a dash dotted line, and acceleration alone as a dotted line. Motion capture is shown in solid black.

	Backbone RMSE (mm)					Drift (mm/50 s)
Method	2D curl	2D obst.	3D curl	3D hand	Avg.	Cyclic	Hold
Fusion (this work)	$7.6\pm 1.1$	$11.3\pm 1.1$	$13.3\pm 0.9$	$7.8\pm 4.0$	$10.0\pm 3.2$	4.1	1.2
Magnetics only	$36.9\pm 2.2$	$18.0\pm 1.5$	$29.4\pm 0.9$	$14.1\pm 2.7$	$24.6\pm 9.5$	51.1	3.3
Constrained acceleration + quaternions	$133.8\pm 3.4$	$163.0\pm 3.4$	$71.9\pm 2.1$	$90.5\pm 4.3$	$114.8\pm 36.8$	54.6	17.1
Acceleration only	$189.0\pm 20.4$	$153.8\pm 15.3$	$177.7\pm 9.1$	$180.8\pm 21.1$	$175.3\pm 20.7$	24.5	0.3
	Tip RMSE (mm)

TABLE I: Position estimation accuracy and reported drift. Scenario RMSEs are mean

\pm

standard deviation against motion capture over five trials; Avg. reports the mean

\pm

standard deviation across all 20 trials pooled across the four scenarios, computed from the rounded scenario statistics. Drift values are changes per 50 s averaged across

x

y

, and

z

during recordings without ground truth and may include physical motion.

Fig. 5: Task demonstrations and long-term estimation. a) Open-loop shirt grasping with grip attachments: (i) initial and (ii) final configurations. b) Height regulation at 40 mm under a 500 g tip load: (i–ii) configurations before and after loading; (iii) height response. Blue and orange dots mark the illustrated frames; the red dashed line marks load application. For this demonstration, we removed the end module to demonstrate that the shape estimation ability is independent of the number of modules. c) Estimated displacement during (i) a stationary hold and (ii) cyclic actuation. The cyclic plot shows $z$ displacement relative to the first cycle.

Work	$L$ [mm]	Sensing / method	Tip error	Shape error
Baaij et al. [ 20 ]	110	3 magnetometers + magnet, NN	—	4.5% (config.)
Adamu et al. [ 21 ]	240	Hall + magnet/segment, MLP	$<$ 5% $L$	max 11 mm per marker
Guo et al. [ 19 ]	—	embedded magnets + sensors	0.23–9.34 mm	—
Li et al. [ 24 ]	—	Hall + base magnet + IMUs	—	$<$ 1 mm
Pittiglio et al. [ 25 ]	64	ball chain + Hall array	2.9% $L$ (max 7.1%)	—
Thuruthel et al. [ 26 ]	120	cPDMS strain, LSTM	$2.4\pm 2.2$ mm	—

TABLE II: Reported proprioceptive sensing accuracy.

L

: robot or sensed length; —: unreported. Metrics and experimental conditions differ across studies. For our work, the row averages scenario means; this work’s shape error is backbone RMSE across nine tracked segments.

Explore similar work

Tendon-Driven Continuum Robot with Modular Stiffness and In-Situ Self Pose Estimation
Sep 14, 2026Guo Ning Sue, Zheng Cao, Junzhe Hu +5Soft Robotics Proprioception
Carnegie Mellon University
Sliding Sensors: Configurable Confidence in State Estimation for Continuum Robots
Aug 5, 2026Ella Walsh, Spencer Teetaert, Eric Diller +2State Estimation Soft Robotics
University of Toronto Robotics Institute, Toronto, ON, Canada
Estimation and Control of Tensegrity Manipulator Kinematics based on Strut Inclination Angles
Sep 21, 2026Tufail Ahmad Bhat, Shuhei IkemotoRobotic Manipulator Soft Robotics
Kyushu Institute of Technology, Japan

Sep 14, 2026, cs.ROJ/K move · Enter open · S save

Tendon-Driven Continuum Robot with Modular Stiffness and In-Situ Self Pose Estimation

Guo Ning Sue, Zheng Cao, Junzhe Hu, Xiangyun Bu+4

Carnegie Mellon University

Soft Robotics Proprioception Robot Systems Uncertain Joint Stiffness Category-Level Object Pose Estimation

Open paper AskPDF arXiv

Continuum robots enable smooth shape morphing and safe interaction in confined environments. However, most existing systems are task-specific and depend on external sensing infrastructure, limiting their adaptability and real-world deployment. This paper presents a self-contained modular continuum robotic platform that combines mechanical reconfigurability with onboard pose estimation. The robot is constructed from interchangeable continuum joints with analytically precomputed stiffness, allowing rapid assembly and direct programming of the robot shape. Proprioceptive sensing is achieved using magnetic sensors and a modular learning-based framework, where a single model is trained per joint and reused across configurations. The system is experimentally validated in real world, demonstrating self-sensing capabilities and adaptation without external tracking.

Abstract

Explore similar work

Tendon-Driven Continuum Robot with Modular Stiffness and In-Situ Self Pose Estimation

Sliding Sensors: Configurable Confidence in State Estimation for Continuum Robots

Estimation and Control of Tensegrity Manipulator Kinematics based on Strut Inclination Angles