Plantar-pressure-guided support-plate position control for a walker-assisted lower-limb rehabilitation exoskeleton Authors: Yuhui Yang, Wenyu Zhao, Yuanxiang Guo, Zhijue Huang, Xueshan Gao, Xin Han, Junlin Deng Published: 2026 ABSTRACT IntroductionMaintaining appropriate foot-ground contact is important for walker-assisted lower-limb exoskeleton training. Excessive plantar loading may increase initial-contact impact and actuator load, whereas excessive unloading may reduce frictional contact and contact stability. Passive spring support can provide impact buffering, but it cannot adjust the support working point according to gait phase or plantar-pressure feedback.MethodsThis study proposes a multi-zone plantar-pressure-guided position control method for the support-force adjustment plate of a walker-assisted lower-limb exoskeleton. The system combines a fixed-stiffness spring with a vertical screw-driven adjustment plate. Total plantar pressure was obtained by fusing signals from 18 pressure cells on each instrumented foot and served as the main feedback variable. Regional plantar pressures and the center of pressure supported gait-phase recognition and contact-state estimation. A phase-related plantar-pressure target band was used as an engineering constraint to maintain stable foot-ground contact. Experiments involved three healthy male participants under no-spring, passive-spring, and proposed-control conditions during low-speed frame-assisted walking.Results and discussionIn the proposed-control condition, peak total plantar pressure was 0.687±0.015 BW, 21.2% lower than in the no-spring condition and 6.7% lower than in the passive-spring condition. Relative to the no-spring condition, the loading rate decreased from 9.02±0.53 to 2.56±0.07 BW/s, and the initial-contact impact impulse decreased from 0.143±0.009 to 0.064±0.002 BW ⋅ s. Hip-knee actuator electrical energy decreased from 38.95±2.20 to 27.24±1.54 J/gait cycle. After vertical-axis actuator energy was included, measured total actuator electrical energy was 30.60±1.68 J/gait cycle and remained lower than in both comparison conditions. These preliminary results suggest that plantar-pressure feedback can regulate support-plate position, attenuate contact impact, and reduce measured robotic actuator effort during low-speed walker-assisted walking. Introduction Lower-limb rehabilitation exoskeletons are designed to provide repetitive, controllable, and quantifiable gait assistance for users with impaired walking function. Recent studies have advanced wearable and mobile lower-limb rehabilitation robots in mechanical design, trajectory tracking, interaction control, and assisted locomotion (Qiu et al., 2023; Jin and Guo, 2023; Yu et al., 2024). For users with weak trunk control or limited independent balance, a walker-assisted architecture provides an additional physical support boundary. This architecture improves operational safety, but it also changes how the human body, the exoskeleton, and the ground exchange load during stance. Therefore, a walker-assisted exoskeleton should be evaluated not only by joint motion tracking, but also by plantar-contact state, vertical support behavior, and actuator effort. Body-weight support has been widely investigated as a way to reduce lower-limb loading during gait training. Cable-driven, mobile, and actively adjusted body-weight support systems can provide unloading and protection during overground or treadmill-based training (Li et al., 2023; Dong et al., 2021; Rodrigues and Goncalves, 2023; Qin et al., 2023; Choi, 2022). These systems show that vertical support can improve the feasibility of gait practice, but they also reveal a control trade-off. If support is purely passive, the spring or suspension can absorb part of the impact but cannot adapt its working point to gait phase or individual differences. If support is actively adjusted without contact-state feedback, the user may experience either insufficient unloading during stance or excessive unloading that weakens foot-ground contact (Park et al., 2022). This motivates a support mechanism that combines fast passive buffering with low-frequency position adjustment driven by plantar-pressure feedback. Plantar pressure provides a direct measurement of the foot-ground contact outcome. The stance-phase plantar-pressure curve has a time-varying trajectory related to gait phase, rather than a single constant value (Wolff et al., 2023; Wolff et al., 2024). Modern pressure insoles commonly use multiple sensing cells, which makes it possible to calculate total plantar pressure, regional pressures, and the center of pressure (COP) after calibration (Ren and Liu, 2021; Conner et al., 2022). These signals have been used for gait-phase recognition, walking-speed estimation, abnormal-gait classification, and multimodal lower-limb state estimation (Liu et al., 2024; Cho, 2023; Jun et al., 2021; Ren et al., 2023). For control purposes, however, a single pressure cell is vulnerable to local contact changes and insole placement. A more robust strategy is to fuse all pressure cells into total plantar pressure while retaining heel, midfoot, forefoot, and COP information to describe load transfer. Multi-sensor fusion and learning-based state estimation have also been introduced into lower-limb exoskeleton control. Integrated network models, compound networks, flexible pressure-sensor arrays, and limb-angle prediction methods have improved gait-phase recognition and lower-limb state estimation (Zhang et al., 2022; Xia et al., 2023; Zhou et al., 2025; Wang et al., 2025). Most of these studies use plantar pressure or inertial signals mainly as recognition inputs, while the outer-loop control objective remains focused on joint motion or joint assistance. In walker-assisted walking, plantar pressure should also be treated as a contact constraint. Excessively low plantar pressure can reduce the observable foot-ground contact state and weaken the pressure-based load-transfer information used for gait interaction (Wolff et al., 2023; Conner et al., 2022; Ren et al., 2023). Therefore, the controller in this study uses a phase-related pressure band, not a maximally unloaded or fixed pressure target. Energy evaluation is another important issue in support-assisted exoskeleton control. Reducing hip and knee actuator load is beneficial only if the additional support actuator does not introduce a larger energy cost. Studies on assist-as-needed control, exoskeleton energy optimization, energy recycling, and measured actuator electrical energy show that actuator-level energy should be reported with a clear definition of voltage/current measurement and integration method (Wang et al., 2022a; Woo et al., 2021; Lee and Rosen, 2023; Andersson and Bjorsell, 2022; Tang et al., 2022; Chen et al., 2025). For the present system, the measured energy metric is limited to the hip, knee, and vertical-axis drive channels and is reported as measured actuator electrical energy. This study therefore treats plantar pressure as a measurable contact constraint for support-plate position control in a walker-assisted exoskeleton. Beyond passive spring buffering, the proposed method adjusts the support working point using phase-related plantar-pressure feedback. The analysis accounts for vertical-axis actuator energy together with hip-knee actuator effort. Materials and Methods describes the robot, sensors, data processing, controller, and experimental protocol. The subsequent sections report plantar-pressure, joint-torque, and actuator-energy results and discuss their engineering implications for low-speed frame-assisted walking. Materials and methods Mechanical structure The robot consists of a walker frame, mobile wheels, a lower-limb exoskeleton body, hip and knee joint drive modules, a foot sensing module, a vertical screw motor, a support-force adjustment plate, and a fixed-stiffness spring (Figure 1). The vertical axis, denoted as the Z-axis in the hardware implementation, follows the direction of adjustment-plate travel. Upward plate motion increases the working compression of the fixed-stiffness spring, whereas downward motion decreases it. The joint modules actively assist hip and knee flexion/extension. The ankle is not actively actuated; the foot module provides mechanical connection and plantar-pressure sensing. One end of the spring connects to the support-force adjustment plate, and the other connects to the back plate or exoskeleton support structure. Spring stiffness remains fixed during operation, so the system is not a variable-stiffness mechanism. Overall mechanical structure of the walker-assisted lower-limb exoskeleton. The figure shows the support frame, cantilever arm, fixed-stiffness spring, thigh module, shank module, foot module, and experimental use scenario. This arrangement provides active-passive hybrid support. The fixed-stiffness spring supplies rapid passive buffering and absorbs part of the impact at foot contact. The vertical screw motor adjusts the support-force adjustment plate at low frequency to change the average unloading-related working point. Because the screw mechanism has limited adjustment speed, the vertical motor is not used to respond directly to high-frequency impact. Instead, plantar-pressure feedback adjusts the low-frequency support working point. This design follows the principle of combining rapid compliant buffering with low-frequency active adjustment in body-weight support and compliant exoskeleton control (Li et al., 2023; Dong et al., 2021; Rodrigues and Goncalves, 2023; Qin et al., 2023; Wang et al., 2022b; Hasan and Dhingra, 2022; Yu et al., 2023; Laubscher et al., 2023). Sensor layout The sensor system includes a multi-zone plantar-pressure sensor, dual encoders in the hip and knee joint modules, hip/knee motor voltage and current measurements, a laser distance sensor for adjustment-plate position, and vertical-axis motor voltage and current measurements. As shown in Figure 2, the laser sensor is fixed to the lower housing and measures the support-force adjustment-plate position zp . Plate velocity is obtained by filtering and differentiating this signal. Each instrumented foot contains an 18-cell distributed flexible thin-film plantar-pressure module; the 18 cells are not divided between the two feet. After calibration, each cell provides a local contact-force measurement. The system sums these signals to obtain total plantar pressure Ffoot and calculates heel pressure Fheel , midfoot pressure Fmid , forefoot pressure Ffore , and COP. Spring support force is not measured directly. Its unloading-related effect is assessed indirectly from total and regional plantar pressure and changes in hip-knee loading (Ren and Liu, 2021; Conner et al., 2022; Liu et al., 2024; Cho, 2023; Jun et al., 2021; Ren et al., 2023). The measured variables and their roles in sensing, control, and evaluation are summarized in Table 1. Prototype assembly and physical sensor layout. The figure marks the plantar-pressure sensor, hip/knee joint modules, laser distance sensor for support-force adjustment-plate position, vertical screw motor, and voltage/current acquisition channels. TABLE 1Sensor configuration and function.Sensor/signalMeasured variablesMain functionMulti-zone plantar-pressure sensor F1 - F18 , Ffoot , Fheel , Fmid , Ffore , COPGait-phase recognition, pressure closed-loop control, impact metrics, and load-transfer analysisHip/knee dual encoders qh , qk , q˙h , q˙k Joint-state estimation and support for torque and energy analysisHip/knee voltage/current Uhip , Ihip , Uknee , Iknee Evaluation of hip-knee power, measured actuator electrical energy, and load changesLaser sensor for adjustment-plate position zp , z˙p Vertical-axis position loop and motion constraint of the support-force adjustment plateVertical-axis voltage/current Uz , Iz Evaluation of additional actuator electrical energy from the vertical support mechanism Sensor configuration and function. The main commercial components are identified to support replicability. The plantar-pressure sensor was a distributed flexible thin-film array acquisition module supplied by Guantuo Electronics (Hefei, Anhui, China). The hip and knee actuators were joint modules supplied by Hangzhou ChengTian Technology Development Co., Ltd. (Hangzhou, China). Adjustment-plate position was measured with a laser infrared ranging sensor module supplied by Jingxun Changtong (Weihai, China). These three components were supplied as integrated modules without separate commercial model designations; therefore, their product categories and suppliers are reported. The vertical actuator was an STM57M integrated stepper-servo motor supplied by NiMotion (Nanjing, China). This configuration measures foot-ground contact, adjustment-plate motion, and electrical energy in each actuator channel. Because the spring’s upper attachment moves with the human-exoskeleton structure, the control law uses measurable plantar pressure rather than an unmeasured spring-force state. Total and regional plantar pressure and COP describe the resulting foot-ground contact. Hip-knee joint torque and measured actuator electrical energy quantify changes in joint loading and drive consumption. These measurements form the control chain from plantar-pressure feedback to adjustment-plate position control and actuator-energy evaluation. Data acquisition and signal processing Signal acquisition, synchronization, and calibration The acquisition system records plantar pressure, joint angles, motor voltage/current, and adjustment-plate position on a common time base. Plantar pressure, hip/knee encoder signals, and all motor voltage/current signals were sampled at 100 Hz. The laser distance signal was sampled at 50 Hz. During online control, the latest valid laser measurement was held on the 100-Hz controller time base until the next sample arrived. For offline analysis and figure generation, this signal was linearly resampled to 100 Hz. The online controller therefore used no future samples or non-causal interpolation. All channels were zeroed before each experiment. The plantar-pressure sensor was calibrated under no-load and known-load conditions. The laser sensor was calibrated linearly using the mechanical travel limits and known displacement points. The calibrated local contact force from pressure cell j is denoted by Fj(t) . Total and regional plantar pressures are obtained by summing the corresponding cells (Ren and Liu, 2021; Conner et al., 2022). Plantar-pressure fusion and COP calculation The total plantar force, regional plantar forces, and COP were calculated using Equations 1–3, respectively. Ffoott=∑j=118Fjt (1) Frt=∑j∈ΩrFjt,r∈heel,mid,fore (2) rCOPt=∑j=118FjtrjFfoott (3) Here, Fj(t) is the calibrated contact force measured by the j -th plantar-pressure cell, with j=1,…,18 . Ffoot(t) denotes the total plantar force of the tested side. Fr(t) denotes the regional plantar force, and Ωr is the set of pressure cells assigned to the region r , where r represents the heel, midfoot, or forefoot region. The vector rj=[xj,yj]T is the planar coordinate of the center of the j -th pressure cell in the foot-fixed insole coordinate system. The origin is located at the midpoint of the posterior edge of the effective sensing area, the x -axis is directed anteriorly along the longitudinal axis of the insole, and the y -axis is directed laterally. The vector rCOP(t)=[xCOP(t),yCOP(t)]T denotes the center of pressure and is calculated only when Ffoot(t) is above the contact threshold. Filtering and gait-phase processing Plantar-pressure signals are used for gait-phase recognition and pressure closed-loop control, so both response speed and noise robustness are required. Each pressure-cell signal is first low-pass filtered, and Ffoot , Fheel , Fmid , Ffore , and COP are then calculated. Gait-phase recognition is based mainly on the total plantar-pressure threshold, pressure change rate, and regional load-transfer features. A rapid increase in heel pressure corresponds to initial contact; combined heel and midfoot loading corresponds to loading response; stable midfoot/forefoot loading corresponds to single support; and decreasing forefoot pressure approaching the threshold corresponds to toe-off (Liu et al., 2024; Ren et al., 2023; Zhang et al., 2022; Xia et al., 2023). The support-force adjustment-plate position is processed with median filtering to suppress occasional jumps, and its velocity is calculated using low-pass filtering and central differentiation. Plate acceleration is used only for motion-safety monitoring and is not used to calculate spring support force. The filtering parameters were selected according to the characteristics of low-speed rehabilitation gait. Plantar-pressure signals were processed using a fourth-order low-pass Butterworth filter with a cutoff frequency of 12 Hz to preserve the main variations during contact and toe-off. Joint-angle signals were processed using an 8-Hz low-pass Butterworth filter, and angular velocity was calculated by numerical differentiation after filtering. Motor-current signals were processed using a 20-Hz low-pass Butterworth filter for torque and electrical-energy analysis. The laser distance signal was processed by a three-point median filter followed by a 6-Hz low-pass Butterworth filter before differentiation. Offline figure generation used zero-phase filtering, whereas real-time control used causal filtering with the same cutoff frequencies to prevent future data from entering the controller (Liu et al., 2024; Cho, 2023; Jun et al., 2021; Ren et al., 2023). Hardware control and data flow Figure 3 summarizes the hardware and data-flow architecture linking the robot, controller, upper-computer software, and processing modules. The 18-channel plantar-pressure insole provides F1 - F18 , and the laser sensor provides zp . The hip/knee modules provide qh , qk , torque-related signals, voltage, and current. The vertical-axis channel supplies voltage/current feedback and receives the position command. These signals reach the STM32 controller through CAN, UART, RS485, or analog acquisition links. The STM32 handles low-level acquisition, communication, and execution of the vertical-axis position loop. MATLAB performs filtering, pressure-zone fusion, gait-phase recognition, target-band calculation, position-command generation, and actuator-energy evaluation. LabVIEW provides the display and logging interface. Accordingly, Figure 3 traces the complete control path from sensor measurements to contact-state variables, target position, and low-frequency vertical-axis adjustment. Hardware control and data-flow architecture of the exoskeleton. The STM32 controller acquires signals from the plantar-pressure insole, laser displacement sensor, hip/knee modules, and voltage/current channels. MATLAB and LabVIEW support filtering, pressure-zone fusion, gait-phase recognition, vertical-axis position-command generation, display, logging, and actuator-energy evaluation. Hip and knee joint torque is obtained from the internal module estimate or approximated from motor current. When the module current constant Kt is available, joint output torque is written as τi=KtIiηi , where ηi denotes the transmission efficiency or calibration coefficient. Electrical energy is calculated from voltage and current measured at each drive channel and is therefore reported as measured actuator electrical energy. The non-negative electrical-power metric Pi(t)=Ui(t)|Ii(t)| is integrated over each gait cycle. Hip, knee, vertical-axis, and measured total actuator electrical energies are reported separately (Wang et al., 2022a; Woo et al., 2021; Lee and Rosen, 2023; Andersson and Bjorsell, 2022; Tang et al., 2022; Chen et al., 2025). Pit=Uit|Iit|,Ei=∫0TPitdt (4) Eact=Ehip+Eknee+Ez (5) In Eqs. Equation 4; Equation 5, i denotes the actuator channel, including the hip, knee, or vertical-axis actuator. Ui(t) and Ii(t) denote the measured drive-channel voltage and current, respectively. The absolute current is used because regenerative braking and driver-side energy recovery are not separately measured. T is the duration of one normalized gait cycle, Ei is the corresponding actuator electrical energy per gait cycle, and Eact is the measured total actuator electrical energy of the recorded hip, knee, and vertical-axis actuator channels. Plantar-pressure-feedback-based support-force adjustment-plate position control Control objectives The controller adjusts the target position of the support-force adjustment plate according to gait phase. Its objective is to keep tested-side total plantar pressure Ffoot within the phase-related target band while reducing initial-contact impact and stance-phase hip-knee loading. Here, Ffoot is the unilateral plantar contact force obtained by summing the calibrated outputs of the 18 cells on the tested side. When both feet are instrumented, the two sides can be calculated and controlled separately. The normalization reference BW is the participant’s body weight msg . It excludes the walker frame, exoskeleton weight borne by the frame, and any external assistance from experimenters. Thus, Ffoot/BW supports relative comparisons across participants and conditions but does not represent the complete vertical force balance of the human-exoskeleton-walker system. During unilateral support or low-speed dynamic load transfer, its peak can reasonably approach or slightly exceed 1. The numerical ranges in Table 2 were defined before the formal A/B/C comparison as engineering design intervals for the low-speed walker-assisted task. They were not tuned using the outcome data. Published body-weight-normalized stance-phase curves provided the phase-dependent shape and approximate reference loading envelope. This envelope included rapid loading after initial contact, sustained support-phase loading, and unloading before toe-off (Wolff et al., 2023; Wolff et al., 2024). A retained plantar-load coefficient of η=0.70 provided the nominal scaling. In a previous body-weight-support experiment, 30% unloading produced plantar forces equal to 68±4 % of unsupported walking, which did not differ significantly from the prescribed 70% retained load (Richter et al., 2021). This evidence informed the overall loading magnitude rather than the exact boundary for each gait phase. TABLE 2Gait-phase-related plantar-pressure target ranges.Gait phaseRecognition basisTarget range of Ffoot/BW SwingTarget foot pressure below the contact threshold0.00–0.05Initial contactPressure rises rapidly and exceeds the contact threshold0.05–0.30Loading responsePressure continues to rise and enters load transfer0.30–0.70Single supportUnilateral pressure dominates and remains relatively stable0.50–0.80Toe-offForefoot pressure decreases and approaches toe-off0.15–0.55 Gait-phase-related plantar-pressure target ranges. The phase-specific limits were then rounded outward to the nearest 0.05 BW to accommodate inter-cycle variability and provide control margin. The contact threshold was 0.05 BW, corresponding to 32.9, 36.8, and 46.6 N for S1, S2, and S3, respectively. Nonzero stance-phase loading was retained to avoid excessive unloading. Wider loading-response and toe-off intervals accommodated transient transfer between plantar regions. This procedure yielded target bands of 0.05–0.30 BW for initial contact, 0.30–0.70 BW for loading response, 0.50–0.80 BW for single support, and 0.15–0.55 BW for toe-off. These values are task-specific engineering constraints rather than universal physiological or clinical reference ranges. The ranges in Table 2 are not intended to reproduce the full vertical ground-reaction-force magnitude of normal overground walking. During normal walking, vertical ground reaction force can exceed body weight, particularly during weight acceptance and late-stance propulsion. In the present task, the walker-assisted exoskeleton and spring-supported structure bear part of the load. The controller therefore seeks to attenuate excessive contact impact without eliminating measurable foot-ground contact. The same target ranges were used for all participants and remained unchanged throughout the formal experiments. Fminϕ≤Ffoott/BW≤Fmaxϕ (6) eFt=0,Fminϕ≤Ffoott/BW≤Fmaxϕ (7) eFt=Ffoott/BW−Fmaxϕ,Ffoott/BW>Fmaxϕ (8) eFt=Ffoott/BW−Fminϕ,Ffoott/BW