RoboSpan: Learning Modular Coordination Policies for High-DoF Whole-Body Mobile Manipulation with Teleoperation

Keqiang Ren; Xuecheng Ye; Hanyu Liu; Zhihua Wang; Tao Jian; Yunsheng Ma; Jiayi Wen; Xinran Li

cs.ROOct 14, 2026

RoboSpan: Learning Modular Coordination Policies for High-DoF Whole-Body Mobile Manipulation with Teleoperation

Authors: Keqiang Ren, Xuecheng Ye, Hanyu Liu, Zhihua Wang, Tao Jian, Yunsheng Ma, Jiayi Wen, Xinran Li,

Organizations: Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment and Technology, School of Mechanical Engineering, Jiangnan University, Wuxi 214122, China.

Abstract

Whole-body mobile manipulation suffers from two fundamental challenges: inherent motion heterogeneity across locomotion, arm, and hand modules, and progressive error accumulation over long-horizon multi-stage tasks. Existing teleoperation platforms also lack integrated dexterous tactile sensing to support coordinated whole-body decision-making. This paper presents RoboSpan, a high-DoF whole-body teleoperation and demonstration platform, and the Hierarchical Modular Framework (HMF) for learning modular coordination policies. HMF functionally decouples the 29-DoF action space into fine hand manipulation, arm pose adjustment, and mobile-base spatial configuration branches, and introduces the Coordinated Modular Embodied Transformer (COMET) with inter-modular cross-attention to enable dynamic information exchange and action coordination. Vision-language model driven hierarchical task decomposition structurally suppresses temporal error propagation, while a full-palm 2D tactile representation encodes distributed contact signals for cross-module tactile empowerment. Experiments on a physical 29-DoF robot demonstrate that the proposed method achieves over 20% higher task success rate than mainstream baselines across a range of long-horizon whole-body tasks, markedly improving stability and robustness in contact-rich manipulation.

Figures & tables

Fig. 1: Overview of the RoboSpan framework for high-DoF whole-body mobile manipulation. RoboSpan integrates whole-body teleoperation, multimodal perception, and modular policy learning, enabling VLM-guided hierarchical task decomposition and coordinated Hand–Arm–Mobile execution for long-horizon manipulation.

Teleop System	Interaction Interface	Bimanual	D-Handl	Elevation	Whole-Body Teleop	Wild
AnyTeleop [ 22 ]	Vision-based teleoperation	$\times$	$\checkmark$	$\times$	$\times$	$\times$
Open-TeleVision [ 18 ]	VR Arm-Hand Teleop	$\checkmark$	$\checkmark$	$\times$	$\times$	$\times$
OPEN TEACH [ 19 ]	VR Arm-Hand Teleop	$\checkmark$	$\checkmark$	$\times$	$\times$	$\times$
Mobile ALOHA [ 20 ]	Leader arms + backdrivable base	$\checkmark$	$\times$	$\checkmark$	$\checkmark$	$\times$
BEHAVIOR Robot Suite [ 21 ]	Twin arms + controllers	$\checkmark$	$\times$	$\times$	$\checkmark$	$\times$
Dexora [ 23 ]	VR Hand Teleop + Exo-Teleop	$\checkmark$	$\checkmark$	$\times$	$\times$	$\times$

TABLE I: COMPARISON OF EXISTING ROBOT TELEOPERATION SYSTEMS

Fig. 2: RoboSpan teleoperation demonstration.

Fig. 3: The dexterous hand and full-palm distributed tactile sensing across diverse objects (I–VIII), together with the corresponding tactile maps.

Fig. 4: Overall network architecture of HMF, which consists of three core modules—Hand, Arm, and Mobile—each with a dedicated encoder/decoder and coupled through cross-policy context fusion (IMCA) under VLM-driven multi-stage decomposition.

Method	Task 1	Task 2	Task 3	Average
ACT	48	40	36	41.3
DP U-Net	40	28	26	31.3
Modular ACT	42	33	30	34.1
COMET	52	46	42	46.7
HMF (Ours)	60	54	50	54.7

TABLE II: Ablation study of task decomposition and modular strategy in the HMF framework (%)

Method	Desktop Organization (3000 time steps)						Trash Cleaning (3000 time steps)
Method	Forward	Grasp Tape	Grasp Bowl	Put	Close	Back	Grasp Cup	Grasp Paper	Move	Lift	Put
ACT [ 33 ]	100%	54%	36%	18%	6%	4%	56%	44%	26%	24%	10%
DP U-Net [ 34 ]	100%	32%	18%	6%	0%	0%	30%	14%	0%	0%	0%
HMF(Ours)	100%	76%	68%	60%	48%	46%	72%	68%	54%	52%	50%

TABLE III: Stage-wise success rate on whole-body mobile manipulation tasks

Fig. 5: Qualitative comparison between ACT (top row) and the proposed HMF (bottom row) on real whole-body mobile manipulation tasks; HMF improves performance by referencing the tactile module for perceptual coordination.

Method	Task 1	Task 2	Task 3	Average
ACT w/o t	40	66	48	51.3
ACT + t	44	70	50	54.7
Modular ACT + t	36	60	44	46.7
COMET (Ours)	48	76	60	61.3

TABLE IV: Ablation of distributed tactile sensing and modular coordination in contact-rich manipulation tasks(%)

Fig. 6: Impact of RoboSpan on task performance. (a) Success rate vs. task duration during data collection. (b) Average success rate at the data-collection and robot-inference stages.

Whole-Body Robotic Manipulation

Whole-Body Robotic Manipulation coordinates a robot’s arms, hands, body, and mobility to manipulate objects while maintaining balance and contact.

See all 88 papers

M3-Tele: A Unified Multimodal Teleoperational Framework for Compliant Whole-Body Mobile Manipulation
Sep 7, 2026Hengxiang Chen, Shenwen Deng, Yujian Ma +3Mobile Manipulation Contact-Rich Robotic Manipulation
Arbeit Gruppe Dexterous Robotics Lab, Shenzhen Technology University, China
A Bilateral Teleoperation Framework for Dexterous Manipulation
Jun 13, 2026Stefano Dalla Gasperina, Dong Ho Kang, Haiyun Zhang +10Dexterous Manipulation Contact-Rich Robotic Manipulation
The University of Texas at Austin, Austin, TX, USA · Sony Group Corporation, Tokyo, Japan · Meta Reality Labs Research, Redmond, WA, USA
X-OP: Cross-Morphology Whole-Body Teleoperation via MPC Retargeting
Jun 6, 2026Jen-Wei Wang, Sarthak Kaingade, Andrea Tagliabue +1Robot Teleoperation Whole-Body Robotic Manipulation
Amazon · University of California, Berkeley

Sep 7, 2026J/K move · Enter open · S save

M3-Tele: A Unified Multimodal Teleoperational Framework for Compliant Whole-Body Mobile Manipulation

Hengxiang Chen, Shenwen Deng, Yujian Ma, Gan Ma+2

Arbeit Gruppe Dexterous Robotics Lab, Shenzhen Technology University, China

Mobile Manipulation Contact-Rich Robotic Manipulation Robot Teleoperation Whole-Body Robotic Manipulation Force-Controlled Robotic Manipulation

Open paper AskPDF arXiv

Executing contact-rich tasks efficiently requires the seamless integration of whole-body coordination and physical compliance regulation. However, existing teleoperation and data-collection frameworks often overlook the joint consideration of multimodal perception and coordinated whole-body operation. This limitation can reduce the efficiency and quality of demonstration collection, thereby affecting the effectiveness of downstream policy learning. In this work, we present M3-Tele: A Unified Multimodal Teleoperational Framework for Compliant Whole-Body Mobile Manipulation, enabling stable physical interaction and capturing aligned visual, tactile, force, and proprioceptive observations during task execution. Extensive experiments demonstrate that the proposed framework significantly improves contact-rich teleoperation performance. The proposed controller reduces the force tracking error from 4.132~~N to 0.346~~N, the contact loss from 2.46 to 0.02 events per trial and the tactile deformation error by 65%. User studies across four mobile manipulation tasks also verify the reliability and usability of the proposed system. Furthermore, Diffusion Policy experiments highlight the value of joint tactile and force sensing.

Abstract

Whole-Body Robotic Manipulation

Explore similar work

M3-Tele: A Unified Multimodal Teleoperational Framework for Compliant Whole-Body Mobile Manipulation

A Bilateral Teleoperation Framework for Dexterous Manipulation

X-OP: Cross-Morphology Whole-Body Teleoperation via MPC Retargeting