Human-Like Redundancy Resolution for Real-Time Bimanual Humanoid Teleoperation

Chao Tan; Ming Shin Cindy Sung; Shuxian Liang; Chaoqun Wan; Zhen Huang; Xian-Sheng Hua

cs.ROOct 12, 2026

Human-Like Redundancy Resolution for Real-Time Bimanual Humanoid Teleoperation

Authors: Chao Tan, Ming Shin Cindy Sung, Shuxian Liang, Chaoqun Wan, Zhen Huang, Xian-Sheng Hua

Organizations: Institute of AI for Engineering, Tongji University

Abstract

Real-time bimanual humanoid teleoperation requires accurate wrist tracking while keeping arm configurations predictable. Because a 7-DoF arm admits multiple elbow configurations for the same wrist pose, wrist-only tracking can cause elbow drift or abrupt configuration changes. Fixed anthropomorphic constraints can stabilize elbow posture, but they can also block the redundant adjustments needed near singular configurations. We propose a human-like redundancy-resolution framework that combines adaptive elbow guidance with singularity-robust wrist tracking. Shoulder-wrist geometric retargeting generates robot-scale wrist targets. An adaptive elbow-offset criterion uses the offset direction to guide posture and scales this guidance by the offset magnitude, so the redundant degree of freedom is progressively released as the human arm straightens. A hierarchical velocity solver prioritizes wrist tracking and projects the adaptive elbow guidance into the wrist-task null space. In a fixed-wrist-target ablation, adaptive guidance reduces mean elbow-direction error by 54.4% compared with removing this guidance. In a workspace-boundary stress test, singularity-robust tracking reduces peak raw joint-velocity demand by 97.9% compared with the undamped pseudoinverse. Real-world bimanual tasks and cross-platform retargeting show applicability across robot morphologies.

Figures & tables

Fig. 1: Human-like redundancy resolution. Wrist-only inverse kinematics leaves the elbow underconstrained, whereas a fixed anthropomorphic objective can become unreliable near full extension. Our method adapts elbow evidence and combines it with singularity-robust wrist tracking.

Fig. 2: Framework of Human-Like Redundancy Resolution for Real-Time Bimanual Humanoid Teleoperation. I) Shoulder–Wrist Geometric Retargeting (SWGR) : XR shoulder, elbow, and wrist keypoints with controller poses are retargeted to robot-scale wrist targets. II) Adaptive Elbow-Offset Anthropomorphic Criterion (AEAC) : the elbow-offset direction guides redundancy with adaptive weighting. III) Singularity-Robust Wrist Tracking : J-PARSE regularizes the stacked wrist Jacobian. IV) Hierarchical Velocity Solver : the primary wrist task is combined with AEAC and posture regulation in the task null space to generate joint commands.

Fig. 3: Static evaluation of GMR, Fourier-Teleop, XR-Teleop, and Ours over 15 poses. (a) Wrist position error

e_{\mathrm{pos}}

; (b) wrist rotation error

e_{o}

; (c) arm-plane error

e_{\mathrm{sa}}

; and (d) joint margin

e_{m}

Fig. 4: Dynamic tracking comparison of GMR, Fourier-Teleop, XR-Teleop, and Ours on elbow, circle, and forward-extension trajectories. (a) Wrist position error

e_{\mathrm{pos}}

; (b) wrist rotation error

e_{o}

; (c) arm-plane error

e_{\mathrm{sa}}

; (d) singularity margin.

Panel of Figure 5 — Fig. 5: Dynamic anthropomorphic behavior. (a) Human and robot arm configurations at selected frames of the elbow-motion sequence. (b) Swivel-angle error trajectories of GMR, Fourier-Teleop, XR-Teleop, and Ours.

Fig. 6: Cross-platform retargeting on Fourier GR-1, JAKA K1, and Unitree H1 using identical parameters.

Panel of Figure 7 — Fig. 7: Dynamic singularity behavior. (a) Forward-extension motion sequence and robot configurations near full extension. (b) Frame-wise normalized singularity margin $\rho_{t}$ after the initial transient.

Panel of Figure 8 — Fig. 8: Real-robot validation. (a) Motion sequence covering neutral, extension, near-singular, recovery, and return phases. (b) Stack Rings: bimanual grasp, transport, and placement. (c) Peel Cucumber: one arm holds the object while the other performs a continuous tool stroke.

Config.	Post./deg $\downarrow$	Pos./mm $\downarrow$	$\rho_{t}/(10^{-3})$ $\uparrow$	$\bar{\ddot{q}}$ /(rad/s 2 ) $\downarrow$
Wrist-only	65.65	5.67	0.991	2.82
Fixed swivel target	13.99	10.27	2.115	168.98
Human-side adaptive	16.88	5.66	2.548	3.02
Combined method	17.23	5.66	2.944	3.20

TABLE I: Forward-extension ablation.

Fig. 9: Anthropomorphic ablation. The proposed AEAC setting reduces elbow-direction error while avoiding the larger motion demand of high gain.

Method	Mean elbow $\downarrow$	P90 elbow $\downarrow$	Position $\downarrow$	$\dot{q}$ $\downarrow$
Ours w/o AEAC	45.0	60.0	0.0	0.00
Ours (default)	20.5	38.0	9.8	0.69
Ours (high gain)	16.3	34.7	24.3	1.38

TABLE II: Anthropomorphic ablation with fixed wrist targets.

Fig. 10: Singularity ablation. The proposed velocity mapping suppresses raw joint-velocity amplification while maintaining bounded position error near the workspace boundary.

Method	Reachable $\downarrow$	Unreachable $\downarrow$	Min. cond. $\uparrow$	Max. $\dot{q}$ $\downarrow$
PINV	0.0	745.2	0.0030	124.23
DLS-low	1.5	390.3	0.0030	6.66
DLS-high	18.7	101.9	0.0267	1.52
Ours	4.5	97.8	0.0030	2.63

TABLE III: Singularity ablation on reachable/unreachable tracking and velocity demand.

Robotic Control

Robotic control studies how machines use sensing and computation to produce reliable movement and interaction in changing environments.

See all 584 papers

Gated Residual Body-Hand Coordination for Whole-Body Humanoid Teleoperation
Sep 16, 2026Ruiming Wu, Shuang Li, Liding Zhang +2Bimanual Robotic Manipulation Robot Teleoperation
Chair of Robotics, Artificial Intelligence and Real-time Systems, Technical University of Munich, Munich, Germany · Agile Robots SE, Munich, Germany
AMBIT: Anticipatory Multimodal Body Recruitment for Bimanual Tracking on a Humanoid
Sep 27, 2026Hanlong Li, Sihan Tan, Takeshi Ashizawa +2Bimanual Robotic Manipulation Humanoid Robot Control
Institute of Science Tokyo, Tokyo, Japan
WristMimic: Full-Body Humanoid Control with Wrist-Guided Manipulation
Jul 7, 2026Wongyun Yu, Youngwoon Kim, Minsu ChoContact-Rich Robotic Manipulation Humanoid Robot Control
Pohang University of Science and Technology (POSTECH) · RLWRLD

Sep 16, 2026J/K move · Enter open · S save

Gated Residual Body-Hand Coordination for Whole-Body Humanoid Teleoperation

Ruiming Wu, Shuang Li, Liding Zhang, Alois Knoll+1

Chair of Robotics, Artificial Intelligence and Real-time Systems, Technical University of Munich, Munich, Germany · Agile Robots SE, Munich, Germany

Bimanual Robotic Manipulation Robot Teleoperation Whole-Body Control Humanoid Motion Tracking Residual Policy Learning

Open paper AskPDF arXiv

Whole-body humanoid teleoperation commonly combines a motion-tracking policy with a separate dexterous-hand retargeter. However, independently generated commands do not explicitly preserve body-hand geometric relations, leading to mismatches in relative wrist poses and fingertip positions during bimanual interaction. We present a gated residual coordination framework that keeps both modules frozen and applies bounded corrections to their outputs. A motion-conditioned action gate allocates correction authority across joint groups, while reference-geometry-dependent reward gates emphasize relevant interaction objectives during training. To establish the nominal body controller on Agile One, we introduce multi-pose morphology calibration that jointly estimates triaxial scales and effector-local offsets, together with staged motion dataset curation for training a SONIC-based tracker. The residual policy uses human motion references, initial commands, and robot proprioception without explicit object or contact observations. In simulation, it reduces wrist and fingertip geometry errors by 39.2-56.3% over direct composition on held-out GRAB motions, while preserving whole-body tracking on AMASS, with success rates of 89.03% without residual coordination and 89.29% with it. Ablations characterize the contributions of reward gating, adaptive correction authority, and separate body and hand correction heads.

Abstract

Robotic Control

Explore similar work

Gated Residual Body-Hand Coordination for Whole-Body Humanoid Teleoperation

AMBIT: Anticipatory Multimodal Body Recruitment for Bimanual Tracking on a Humanoid

WristMimic: Full-Body Humanoid Control with Wrist-Guided Manipulation