A Flow-Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper-Limb Rehabilitation

Jun Leng; Pengcheng Li; Hanze Wang; Shuxiang Guo; Junlin Jiang

cs.ROOct 15, 2026

A Flow-Theoretic Multimodal Control Framework Fusing Mental and Physical Engagement for Upper-Limb Rehabilitation

Authors: Jun Leng, Pengcheng Li, Hanze Wang, Shuxiang Guo, Junlin Jiang

Organizations: Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen 518055, China · Aerospace Center Hospital, School of Life Science, and the Key Laboratory of Convergence Medical Engineering System and Healthcare Technology, Ministry of Industry and Information Technology, Beijing Institute of Technology, Beijing 100081, China

Abstract

While Assist-as-Needed (AAN) control aims to maximize user participation in robot-assisted training, existing methods relying solely on physical metrics cannot distinguish muscular inability from mental disengagement, leading to suboptimal interventions. To address this ambiguity, a Flow-theoretic multimodal AAN framework is proposed, fusing EEG-based mental engagement and sEMG-based physical engagement into an adaptive impedance control strategy. A four-quadrant fuzzy decision engine translates these dual physiological metrics into continuous control policies. This enables the upper-limb exoskeleton to inject feedforward assistance during intent-execution mismatch (e.g., motor output attenuation despite mental focus) and escalate guiding stiffness to penalize mental distraction. The framework was validated with healthy participants against a state-of-practice sEMG-only AAN baseline. Kinematically and mechanically, the proposed framework reduced robotic energy expenditure by 71.9% while maintaining precise trajectory tracking, confirming it elicited greater voluntary human effort. Physiologically, the system increased the time spent in the neuroplasticity zone (Flow State Ratio) from 4.6% to 42.8%. These results validate that decoupling and jointly regulating mental and physical states enhances the selectivity of robotic assistance, mitigating user passivity and sustaining the engagement crucial for motor learning.

Figures & tables

Panel of Figure 1 — Fig. 1: System overview and hardware architecture. (a) Posterior view showing the backpack-mounted actuation unit and the Bowden-cable transmission. (b) Lateral view of the user wearing the CURE exoskeleton, detailing the 1-DoF elbow joint and IMU sensor placements. (c) Schematic of the multimodal sensing interface (EEG and sEMG) and the closed-loop data flow architecture.

Fig. 2: Overall architecture of the proposed flow-grounded AAN framework. (a) High-Level Controller: A multimodal decision-making engine that processes EEG and sEMG signals to map the user’s mental and physical engagement into continuous strategic policies via fuzzy inference. (b) Low-Level Controller: An adaptive execution module that synthesizes these policies into targeted impedance parameters ( $K,D,\tau_{\text{assist}}$ ) to regulate sensorless human-robot interaction.

Panel of Figure 3 — Fig. 3: Continuous control output surfaces generated by the Fuzzy Inference System. (a) The Assist Level ( $c^{*}_{\text{assist}}$ ) surface mapped across the two-dimensional physiological state space. (b) The Challenge Level ( $c^{*}_{\text{challenge}}$ ) surface.

Panel of Figure 4 — Fig. 4: Experimental setup and the visual feedback interface. (a) Overview of the experimental scenario during the active tracking task. (b) A representative screenshot of the custom graphical user interface (GUI), illustrating the target trajectory, the user’s real-time cursor, and the synchronized phase prompts.

Panel of Figure 5 — Fig. 5: Experimental procedure and protocol design. (a) Individualized calibration phase. (b) Study I: Continuous adaptive response test (240-s). (c) Study II: Comparative efficacy analysis against a baseline AAN mode (5-min).

Physiological Metric	Induced “Low” State	Induced “High” State	Selected Inflection ( $c$ )
Mental Engagement ( $s^{*}_{\text{me}}$ )	$0.22\pm 0.08$	$0.65\pm 0.12$	0.40
Physical Engagement ( $s^{*}_{\text{pe}}$ )	$0.15\pm 0.06$	$0.58\pm 0.10$	0.30

TABLE I: Physiological Responses During the Isolated 1-Minute Induction Blocks (Mean

\pm

SD)

Panel of Figure 7 — Fig. 6: System dynamic response across the four quadrants during a continuous 240-s training session. The top ribbon indicates the real-time decoded macroscopic states ( $Z_{k}$ ). (a) Smoothed physiological inputs: mental engagement ( $s^{*}_{\text{me}}$ ) and physical engagement ( $s^{*}_{\text{pe}}$ ). (b) High-level continuous policy variables: assistance demand ( $c^{*}_{\text{assist}}$ ) and challenge level ( $c^{*}_{\text{challenge}}$ ). (c) Low-level modulated impedance parameters: virtual stiffness ( $K$ ) and feedforward assistance force ( $F_{\text{assist}}$ ). (d) Kinematic tracking performance.

Phase (State)	$\mathbf{s^{*}_{\text{me}}}$	$\mathbf{s^{*}_{\text{pe}}}$	Stiffness $K$ (N $\cdot$ m/rad)	Assist. Force $F_{\text{assist}}$ (N)
1 (Flow)	High	High	High ( $\sim 10$ )	Minimal ( $<1.0$ )
2 (Boredom)	Low	High	Firm ( $\sim 8$ )	Withheld ( $<1.0$ )
3 (Apathy)	Low	Low	Low ( $\sim 2$ )	Moderate Peaks ( $\sim 2.0$ )
4 (Anxiety)	High	Low	Fluctuating ( $\sim 5$ )	Strong Peaks ( $\sim 2.3$ )

TABLE II: Summary of System Responses Across the Four Induced States

Panel of Figure 9 — Fig. 7: Quantitative analysis of transition smoothness across 23 major inter-quadrant shifts. (a) Boxplot of the controller response latency. (b) Boxplot of the maximum rate of change of the assistance force ( $|dF_{\text{assist}}/dt|$ ).

Evaluation Metric	Baseline AAN	Fusion Mode (FM)	p -value
Tracking RMSE (rad)	$0.177$	$\mathbf{0.174}$	$>0.05$ (n.s.)
Mean Assistive Force (N)	$1.85$	$\mathbf{0.51}$	$<0.001$
Mental Engagement ( $s^{*}_{\text{me}}$ )	$0.35$	$\mathbf{0.45}$	$<0.01$
Physical Engagement ( $s^{*}_{\text{pe}}$ )	$0.21$	$\mathbf{0.34}$	$<0.001$
Flow State Ratio (%)	$4.6$	$\mathbf{42.8}$	$<0.001$

TABLE III: Quantitative Comparison Between Baseline AAN and Proposed Fusion Mode (FM)

Panel of Figure 11 — Fig. 8: Kinematic tracking performance and robotic assistance evaluation. (a) A representative tracking trajectory segment under AAN and Fusion Mode (FM). (b) Grand-averaged Root Mean Square Error (RMSE) across all subjects ( $p>0.05$ ). (c) Grand-averaged mean feedforward assistive force showcasing a significant reduction in robotic physical intervention ( $p<0.001$ ).

Panel of Figure 12 — Fig. 9: Physiological state maintenance and 2D state distribution derived from offline analysis. (a) Grand-averaged mental and physical engagement scores, alongside the Flow State Ratio. (b) 2D Kernel Density Estimation (KDE) map of user states under the baseline AAN mode. (c) 2D KDE map of user states under the proposed Fusion Mode (FM).

Panel of Figure 13 — Fig. 10: Microscopic case studies illustrating the closed-loop autonomous recovery mechanisms. (a) Recovery from Anxiety (intent-execution mismatch): the system decreases virtual stiffness ( $K$ ) and elevates feedforward assistance ( $F_{\text{assist}}$ ) in response to acute motor output attenuation. (b) Recovery from Boredom (attentional drift): the system escalates virtual stiffness ( $K$ ) to apply a haptic penalty in response to dropping mental engagement, thereby prompting rapid cortical re-engagement.

Assistive Robotics

Assistive robotics develops robots and wearable devices that support mobility, daily activities, rehabilitation, and care.

See all 208 papers

Real-Time Decoding of Movement Onset and Offset for Brain-Controlled Rehabilitation Exoskeleton
Mar 17, 2026Kanishka Mitra, Satyam Kumar, Frigyes Samuel Racz +3Brain-Computer Interfaces Assistive Robotics
Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, USA. · Chandra Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX, USA. · Department of Neurology, The University of Texas at Austin, Austin, TX, USA. +3
Attribution-Gated Joint Estimation of Intention and Capability for Assist-as-Needed Upper-Limb Robots
Oct 15, 2026Pengcheng Li, Shuxiang Guo, Hanze Wang +3Assistive Robotics Continuous Control
Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen, China · Aerospace Center Hospital, School of Life Science, and the Key Laboratory of Convergence Medical Engineering System and Healthcare Technology, Ministry of Industry and Information Technology, Beijing Institute of Technology, Beijing, China · Department of Peripheral Vascular Intervention, Aerospace Center Hospital, School of Life Science, Beijing Institute of Technology, Beijing, 100081, China
ATP: Anatomical Torque with Passivity-based Control Framework for Safe Upper-Limb Exoskeleton Assistance
Aug 6, 2026Yu Chen, Gong Chen, Xiang LiAnkle Exoskeleton Control Feedback Control
Department of Automation, Tsinghua University, China · Shenzhen MileBot Robotics Co., Ltd, China

Mar 17, 2026J/K move · Enter open · S save

Real-Time Decoding of Movement Onset and Offset for Brain-Controlled Rehabilitation Exoskeleton

Kanishka Mitra, Satyam Kumar, Frigyes Samuel Racz, Deland Liu+2

Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, USA. · Chandra Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX, USA. · Department of Neurology, The University of Texas at Austin, Austin, TX, USA.+3

Brain-Computer Interfaces Assistive Robotics MI Classification Electroencephalography Upper-Limb Exoskeletons

Open paper AskPDF arXiv

Robot-assisted therapy can deliver high-dose, task-specific training after neurologic injury, but most systems act primarily at the limb level-engaging the impaired neural circuits only indirectly-which remains a key barrier to truly contingent, neuroplasticity-targeted rehabilitation. We address this gap by implementing online, dual-state motor imagery control of an upper-limb exoskeleton, enabling goal-directed reaches to be both initiated and terminated directly from non-invasive EEG. Eight participants used EEG to initiate assistance and then volitionally halt the robot mid-trajectory. Across two online sessions, group-mean hit rates were 61.5% for onset and 64.5% for offset, demonstrating reliable start-stop command delivery despite instrumental noise and passive arm motion. Methodologically, we reveal a systematic, class-driven bias induced by common task-based recentering using an asymmetric margin diagnostic, and we introduce a class-agnostic fixation-based recentering method that tracks drift without sampling command classes while preserving class geometry. This substantially improves threshold-free separability (AUC gains: onset +56%, p = 0.0117; offset +34%, p = 0.0251) and reduces bias within and across days. Together, these results help bridge offline decoding and practical, intention-driven start-stop control of a rehabilitation exoskeleton, enabling precisely timed, contingent assistance aligned with neuroplasticity goals while supporting future clinical translation.

Abstract

Assistive Robotics

Explore similar work

Real-Time Decoding of Movement Onset and Offset for Brain-Controlled Rehabilitation Exoskeleton

Attribution-Gated Joint Estimation of Intention and Capability for Assist-as-Needed Upper-Limb Robots

ATP: Anatomical Torque with Passivity-based Control Framework for Safe Upper-Limb Exoskeleton Assistance