Anthropometric measurements were collected from 20 male participants (average age 27.5, SD 2.76 years) to establish design parameters for the modular wheelchair. This demographic was selected to determine conservative upper-bound dimensions, recognizing that male participants generally have larger anthropometric dimensions. Measurements were taken with the user seated in an optimal position as shown in Figure 1. The dimensions collected included hip width, buttock-popliteal length, popliteal height, subscapular height, shoulder height, elbow height, knee height, shoulder width, elbow fingertip length, upper limb length, and shoulder grip length. These measurements were used to develop a fit and reach matrix according to user comfort, with 5th and 95th percentile values calculated for design purposes (Table 1).

The wheelchair design dimensions were based on standards and guidelines from the Central Public Works Department (CPWD), the Office of Chief Commissioner for Persons with Disabilities, IS 4963 (building code standard), and IS 7454 (wheelchair specification standard). The clearance dimensions are summarized in Table 2.

The wheelchair dimensions for clearances and reach were based on the maximum of maximum (95th percentile) and minimum of minimum (5th percentile) human anthropometric dimensions, respectively. Wheelchair dimensions such as seat length and backrest angle were designed to be adjustable over the 5th and 95th percentile range using an angle-adjustable backrest and a slide-lock mechanism. The reach space design was based on wheelchair height (H1) for vertical reaches and length L1 as shown in Figure 2A-2B, width L2 (Figure 2B), and distance L3 (Figure 2C) for horizontal distances, calculated using a digital human model (DHM).

The primary reach relationships derived from the DHM are given by equations (1–4):

Forward high: (1)

Forward low: (2)

Lateral high:    (3)

Lateral low: (4)

Where H1 is the height of the wheelchair from the ground to the lowest point on the seat (500 mm), E is the shoulder height from the seat, λ is the torso angle the participants make with the coronal plane, K is the shoulder grip length, and τ is the angle the torso makes with the sagittal plane.

The minimum turning distance required for a 360° turn was calculated using Equation (5):

Tcircular=2×π ×L12+L22     (5)

Where L12+L22 is the turning radius of the wheelchair

For this prototype, L1=600 mm and L2=950 mm

Therefore, the distance required for a 360^o turn calculated using DHM will be,

T_circular=4870 mm.

The manual stair-climbing wheelchair is a dual-purpose wheelchair designed to provide multidimensional accessibility on both stairs and plane surfaces. To test the efficacy of wheelchair driving on plane surfaces, a simulated course was designed to test maneuverability and control. The simulated course contained architectural hurdles that a wheelchair user faces in an urban setting, including a 5° ramp followed by a sharp 90° turn and then converging and diverging spaces. The course is illustrated in Figure 3.

This pilot study used n=9 participants for plane surface assessment (7 healthy, 2 with paraplegia), following established guidelines for feasibility studies in assistive technology research [12,13]. Two participants with paraplegia provided critical real-world user validation. The course was designed as a controlled experimental environment to ensure participant safety and enable systematic data collection, including multiple stair configurations with varying heights (15‐20 cm) and depths (25‐30 cm), curved pathways, simulated obstacles, and various surface transitions representative of typical urban environments.

The subjective descriptors used for the response for each riding task using Likert scales were (1) strongly favorable (score=4), (2) moderately favorable (score =3), (3) moderately unfavorable (score=2), and (4) strongly unfavorable (score=1). Task completion times were recorded for each of the 9 tasks across 3 trials. After the completion of all tasks, a subjective survey of the whole task was taken on a visual satisfaction scale ranging from 0 (Very negative) to 10 (Very positive) for overall ride comfort, stability, driving control, perception, and maneuverability.

The biomechanical evaluation of arm and shoulder muscles was performed to identify and mitigate shoulder pain and fatigue generated during lever propulsion. Use of sEMG is a prominent method for recording muscle activation artifacts to understand muscle effort, muscle activation timing, and muscle fatigue. The activation timing study of a muscle group was performed to understand the loading profile of muscle at a specific interval in the work cycle [14]. Surface electromyography also helps in identifying muscles under fatigue by observing an increase in amplitude and decrease in median frequency of the bioelectrical signal [15].

The upper limb and shoulder are the most common sites for any possible stress injury or musculoskeletal disorder due to long-term cyclic loading during lever-based propulsion [16,17]. The muscle groups under study for this specific research go under cyclic dynamic interactions and muscle fiber contractions during the lever-based propulsion cycle and are anisometric (concentric and eccentric) in nature. Therefore, the shoulder and arm joint angles are also required to be recorded while recording sEMG activity. The recorded electromyography signal has artifacts from the velocity- and angle-dependent muscle excitation of the shoulder and elbow extensors and flexors. The muscle activation of the extensor and flexor groups will be significant during both actions of pulling and pushing. But, since only pulling stroke is responsible for stair climbing, only those muscles will show significant dynamic electromyography activity that are associated with pulling action (elbow flexors and shoulder extensors muscle groups).

The action of lever propulsion in wheelchairs and its physiological effects are reported by only a few researchers, as the wheelchairs developed with a lever mechanism are very much limited, and in consequence, electromyography-based ergonomic evaluations based on them. There are few studies about lever-based propulsion effort, but they are limited to physiological measurements such as oxygen uptake, respiration ratio, minute ventilation, heart rate, etc [18-20]. With the advent of surface electromyography recording and processing techniques, the idea of using electromyography as a dominant physiological parameter that can influence design is increasing. Push-rim–based propulsion of a plane-surface wheelchair has been highly explored using surface electromyography studies on dominant-side muscle groups such as the deltoid group, triceps group, and bicep groups. As reported in the literature, there are some overlapping muscle groups engaged during both lever propulsion and hand-rim propulsion, such as biceps brachii, posterior deltoid, triceps, pectoralis major, etc [21-23]

The action of lever propulsion is a push-pull action that requires elbow (flexion and extension) and shoulder (flexion and extension) movement. The action of pulling requires flexion of the elbow and extension of the shoulder muscles and vice versa for pushing. The pulling stroke is responsible for rotating the stair-climbing wheel on the staircase. The pushing stroke (moving the lever away from the chest) is idle and requires less effort compared to pulling. Tomiak et al [24], in their research related to bimanual rowing movement, which mimics a lever push-pull activity, group together the pulling and pushing (returning) muscle groups as elbow and shoulder muscles. The primary goal of the ergonomic evaluation in this section is to identify extreme joint loading forces by trying different posture configurations defined by ergonomic factors. Pushing and pulling tasks can cause extreme lower back and upper joint loading, leading to musculoskeletal disorders (MSDs) [25,26].

Four muscle groups were selected for sEMG recording: (1) biceps brachii long head (BBL), a primary muscle involved in flexing the elbow, selected due to availability of higher density of motor neurons (Figure 4A); (2) triceps brachii long head (TBL), located on the back of the upper arm, responsible for extending the elbow during the pushing stroke (Figure 4B); (3) brachioradialis, responsible for forearm and elbow flexion necessary for pulling, with secondary function of supination. and pronation for holding the lever (Figure 4B); and (4) posterior deltoid (PDT), located at the shoulder, extends during shoulder extension required for pulling the lever toward the chest (Figure 4C).

Three ergonomic factors were evaluated in this study:

(1) Lever orientation (ψ): the range of motion for supination and pronation was limited to ±30°, where supination is +30° (Figure 5A), the neutral position is 0° (Figure 5B), and pronation is −30° (Figure 5C). Although Sullivan et al [27] reported that up to a certain degree of pronation, subjective discomfort is lesser and higher effort can be achieved compared to supination and neutral hand orientation [28], objective evaluation using electromyography was required.

(2) Torso angle (λ): 3 trunk orientations in the coronal plane were studied [29,30], including 45° (Figure 6A), 30° (Figure 6B), and 0° upright (Figure 6C). The maximum low reach can be achieved at 45°, while maximum stability and minimum spinal loading can be attained at 0°.

(3) Lever distance (L): the distance from the shoulder to the lever (shoulder grip length+L) was varied using a novel seat adjustment mechanism. The distance L was calculated from the lower back of the seated participant to the backrest of the wheelchair. Three levels were tested, including L=0 mm (Figure 7A), L=50 mm (Figure 7B), and L=100 mm (Figure 7C).

Seven healthy participants (S1–S7) were selected for the surface electromyographic study. The anthropometric data of the participants is shown in Table 3.

A 3-factor, 3-level experimental design was used. The factors and their levels are shown in Table 4. A full factorial design would require 27 runs; this was reduced to a 1/3 fractional factorial design using the Taguchi method, requiring 9 experimental runs (Table 5). Each participant performed the lever push-pull action while seated in the configuration specified by each experiment.

The test wheelchair used in this study is a manually propelled stair-climbing wheelchair with a ratchet-based lever mechanism that provides a to-and-fro motion where pulling toward the chest engages the lever with the propulsion shaft, providing rotational motion to the stair-climbing wheel. For recording sEMG data, an ADS1299-based evaluation board (Texas Instruments) was used, providing 8 analog channels with a common reference, as reported in earlier sEMG studies [31-33]. These 8 channels were configured to create 4 channels with a common-mode rejection ratio for high-quality sEMG data for all 4 muscle groups as shown in Figure 8. An instrumentation gain of 11X and an overall system gain of 2420X were applied. Ultra-low impedance (<100 ohms) nickel-chromium (Ni-Cr) alloy electrodes were used with silver-silver chloride electrode (Ag-Cl) adhesive solid hydrogel at the electrode-skin interface to reduce skin impedance. The sampling frequency was 500 Hz.

The raw electromyography signal was filtered using a band-pass Butterworth Infinite Impulse Response filter. The filtered electromyography signal was further processed by taking the absolute value and applying a moving average filter (window of 50 samples) as shown in Figure 9. For identification of muscle activation timing based on adaptive threshold, the Hilbert transform was used with a threshold duration of 25 samples to avoid influence of noise or other involuntary artifacts visible in Figure 10A. From the activation timing study, an activation window for the pull-push cycle is illustrated in Figure 10B.

Maximum voluntary contraction (MVC) was calculated from the digitally filtered electromyography envelope with absolute values as a quantifiable parameter for normalizing the electromyography signal. The weighted average of envelope amplitude and envelope root-mean square (RMS) was used to calculate MVC [34], expressed as:

MVC=w1 ✕ A+w2 ✕ RMS        (6)

Where,A is the maximum amplitude of the electromyography envelope given by

A=max (X) –min (X)         (7)

 X={x1, x2, x3……xn}; X is the dataset of filtered electromyography signal recorded over time t.

x_i represents the amplitude of the electromyography signal at i-th time point.

RMS means root mean square of dataset X.

Following Merletti and Parker (2004) [34], the weights were set as w1=0.7 and w2=0.3, prioritizing sensitivity to maximum activation for detecting full recruitment peaks while complementing it with RMS stability to mitigate outlier effects.

Muscle coordination patterns were assessed using the Pearson correlation coefficient “r” to measure the linear relationship between MVC values of muscle pairs [35]. The coefficient is given by:

where x_i and y_i are the MVC values for 2 muscles (eg, anterior deltoid and biceps brachii) across all observations, and x- and y- are their means.

Quadratic regression models were fitted for each ergonomic factor–muscle response relationship to characterize the design space. Main effects and interaction effects (AB: λ×L; AC: λ×ψ; BC: L×ψ) were computed by averaging MVC values across paired factor combinations. Interaction effects were visualized using line plots and heatmaps. 3D surface plots were generated to delineate the ergonomic design space. Statistical significance of correlations between anthropometric dimensions and task performance was assessed at P<.05.

The time taken for each of the 9 participants to complete each task on the simulated course is presented in Table 6.

The ramp (R) and sharp 90° turn (T1) emerged as the most time-intensive tasks. The ramp required an average of approximately 14.8 seconds, while the 90° turn required approximately 21.6 seconds. Tasks on smooth floor surfaces (F1 and F2) and rubber carpets (R1 and R2) required relatively less time, averaging 2‐5 seconds each.

The time required to complete the 9 standardized wheelchair driving tasks, along with participant favorability ratings, is summarized below. All times are reported as mean (SE) in seconds.

Task C (straight corridor): Participants completed the task in 4.1 ± 0.9 s and rated it as moderately favorable. Task R (right turn): 14.8 (1.8)s, rated as moderately unfavorable. Task T1 (tight turn 1): 22.3 (2.1) s, rated as strongly unfavorable. Task F1 (forward maneuver 1): 5.7 (1.2) s, rated as moderately unfavorable. Task R1 (reverse 1): 2.4 (0.4)s, rated as moderately favorable. Task N (narrow passage): 8.5 (1.5) s, rated as strongly favorable. Task F2 (forward maneuver 2): 2.5 (0.5) s, rated as moderately favorable. Task R2 (reverse 2): 4.1 (0.8) s, rated as moderately favorable. Task T2 (tight turn 2): 7.8 (1.1) s, rated as moderately favorable.

These results indicate that tasks involving tight maneuvers (particularly T1) required substantially more time and received the least favorable ratings, whereas simpler forward and reverse tasks were completed quickly and viewed more positively. The pattern highlights the importance of adequate space and clear pathways in wheelchair-accessible environments.

In the overall subjective response study for all 9 participants evaluating wheelchair attributes on a visual analog scale from 1 to 10, overall ride comfort received the lowest mean rating of 3.33 (SD 1.58), indicating areas for potential improvement in user experience during mobility. Wheelchair stability was rated moderately higher at a mean of 5.44 (SD 1.74), while driving control emerged as the strongest attribute with the highest average score of 6.00 (SD 1.22), suggesting effective handling and responsiveness in the device's design. Perception of the wheelchair scored a mean of 3.89 (SD 1.17), reflecting possibly mixed sensory feedback, whereas maneuverability—potentially encompassing response aspects—achieved a solid mean of 5.11 (SD 1.17), highlighting reliable navigation capabilities as depicted in the bar graph with error bars representing SDs. The correlation between task difficulty (higher time and low level of easiness) and anthropometric dimensions of the participants was not statistically significant ( P=.04), indicating that task difficulty could be directly associated with wheelchair design variables rather than participant characteristics.

The average electromyography amplitude and RMS of all 7 partiticipants for each muscle group, along with calculated MVC values, are presented in Table 7. There was no significant statistical correlation between the anthropometric features and electromyography activity in terms of MVC (P=.04). The maximum electromyography amplitude for each muscle averaged from all 9 experiments is illustrated in Figure 12A. Muscle coordination patterns derived from MVC data using the Pearson correlation coefficient are shown in Figure 12B.

Quadratic regression models were fitted for each factor-muscle relationship. The regression equations, residual sum of squares, and coefficients of determination are presented in Table 8.

The main effects of the 3 ergonomic factors on MVC for each muscle group are summarized in Table 9 and visualized in the main effect plots (Figure 12).

For torso angle (λ), MVC increased with inclination across all muscles, including BBL from 38.46 to 44.78, TBL from 47.96 to 67.49, brachioradialis from 72.16 to 113.41, and PDT from 50.51 to 77.77, with a straight torso (0°) yielding the lowest strain across all muscles, particularly brachioradialis. For lever distance (L), BBL remained relatively stable (40.47‐42.57, lowest at 50 mm), TBL was optimized at 51.30 (50 mm), brachioradialis rose from 89.46 to 97.78 with decreasing L, and PDT was lowest at 59.67 (50 mm). For lever orientation (ψ), BBL slightly favored 30° (40.79); TBL and PDT were minimized at 0° (53.11 and 59.58, respectively); and brachioradialis peaked at 0° (106.65) but dropped to 86.11 at 30°.

Interaction effects (AB: λ×L; AC: λ×ψ; BC: L×ψ) were computed by averaging MVC values across paired factor combinations. These interactions are visualized in line plots (Figure 13) and heatmaps (Figure 14).

For the AB interaction (Torso angle ×Lever distance), forward lean (λ=45°) combined with long lever distance (L>50 mm) produced elevated effort in BBL and TBL, with lines crossing in Figure 13 and heatmaps showing high-strain zones. Upright posture (λ=0°) at L≈50 mm maintained low effort levels.

For the AC interaction (Torso angle ×Lever orientation), supination (ψ =+30°) combined with λ=0° reduced effort for brachioradialis and PDT, while pronation (ψ = –30°) at λ=45° increased effort, shown by crossing lines and heatmap contrasts. TBL was less affected by this interaction.

For the BC interaction (Lever distance ×Lever orientation), L ≈ 50 mm with ψ =+30° minimized brachioradialis and PDT strain, while a long lever distance with ψ = –30° maximized strain. TBL effort increased with pronation (see Figure 14).

The design space heatmaps (Figure 15) and their associated 3D surface plots (Figure 16) delineated the ergonomic design space. For the AB interaction (λ×L), heatmaps revealed BBL (MVC 38‐50) and PDT (50-95) with high strain at λ=45°, L=100 mm, and low strain at λ=0°, L=0 mm. The 3D surface plot peaked at 80 MVC where λ=45° and L=100 mm and dipped to 50 MVC at λ=0° and L=0 mm as shown in Figure 16A.

For the AC interaction (λ × ψ), BBL minimized strain at λ=0°, ψ=30° (supination), and PDT at λ=0°, ψ=0°, with high strain at λ=45° and ψ=−30°. The 3D plot rose to 75 MVC at λ=45°, ψ=−30° and fell to 50 MVC across λ=0° and ψ=0°–30° as shown in Figure 16B. For the BC interaction (L × ψ), BBL was optimized at L=50 mm, ψ=30°, and PDT at L=50 mm, ψ=0°. The 3D plot peaked at 80 MVC for L=100 mm, ψ=−30° and reached approximately 50 MVC near L=50 mm, ψ=0°–30°, as shown in Figure 16C.

Collectively, strain exceeded 75 MVC at extreme configurations (eg, λ=45°, L=100 mm, ψ=-30°) and dropped to approximately 50 MVC at optimal settings (eg, λ=0°, L=0‐50 mm, ψ=0°-30°).

Optimal settings for minimizing individual muscle activity are shown in Table 10. These preferences vary by muscle but consistently favor a straight torso.

Averaging across all muscles for a generalized design, the overall ergonomic recommendation is presented in Table 11 along with optimal wheelchair design parameters in Table 12.