FES-Cycling at the Cybathlon 2024: assessing accuracy of power-pedals for low power output in a virtual race scenario

Abstract

Background

The Cybathlon is a unique international competition platform for evaluating assistive technologies under real-life conditions. In the most recent edition, the FES-Cycling discipline was conducted on stationary smart trainers (Wahoo Kickr) within a virtual race environment. Participants could use either the smart trainer or a pedal-based power meter as the primary power source. As both systems were not specifically designed for the very low power outputs typically generated during FES-Cycling, the outcomes of the competition require careful interpretation.

Methods

This work summarizes the FES-Cycling competition at the Cybathlon 2024 and evaluates the accuracy of pedal-based power meters used in the virtual race. Race performance data from ten pilots representing eight international teams were analyzed. Pedal-based power measurements were compared against a validated smart trainer during the competition. In addition, an assessment in a controlled environment was conducted to quantify pedal accuracy in the low-power range (< 25 W) using a crank-based power meter as reference. The influence of pedal mounting torque, ankle-orthosis mounting position, and crank-length compensation was systematically investigated.

Results

Four of the ten pilots completed the full virtual race distance of 1960 m within the 8-min time limit. Mean cycling speeds (7.8–19.5 km/h) exceeded those reported for the over-ground Cybathlon 2016 race. Analysis of valid race segments revealed substantial overestimation of pedal-based power relative to the smart trainer, with relative errors ranging from 14.7 to 66.6%. Additional accuracy measurements confirmed that pedal-based power was sensitive to mounting torque and particularly to ankle-orthosis position, with lateral displacement of the foot causing pronounced overestimation. Crank-length compensation reduced systematic bias, but residual errors and increased variability persisted at higher cadences.

Conclusion

Commercially available pedal-based power meters and stationary smart trainers are currently not well suited for reliable performance assessment in low-power FES-Cycling competitions within virtual race environments. While virtual formats offer clear logistical advantages, engaging training opportunities and rich performance data, traditional race formats should be reconsidered to better align with the functional goals of FES-Cycling. Future competitions should prioritize over-ground cycling, assessing externally observable outcomes such as distance and time, which may further enhance pilot motivation and public visibility.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12984-026-02025-3.

Background

Since its initiation in Zurich in 2016, the Cybathlon has served as a facilitating platform across multiple disciplines within the broader field of assistive technologies [1]. By introducing competitive elements, the Cybathlon has shifted the focus from purely clinical applications toward innovation and technological advancement, clearly demonstrating its impact on the development of novel assistive solutions [2].

Functional electrical stimulation (FES) cycling [3] was among the core disciplines introduced in the first Cybathlon and was open for pilots with motor-complete spinal cord injury (SCI). In 2016, eleven international pilots with SCI were challenged to complete an over-ground distance of 742 m [4] within a time limit of eight minutes. Seven pilots successfully covered the full distance, six of them using surface stimulation electrodes and one pilot employing a fully implanted stimulation system. The average speeds of pilots using surface stimulation ranged from 5.7 km/h to 11.2 km/h, while the pilot with the implanted system achieved an average speed of 15 km/h. These results remain benchmark performances for over-ground FES-Cycling to this day.

In the second edition of the Cybathlon in 2020, it was not possible to conduct the FES-Cycling race in person due to the COVID-19 pandemic. Consequently, the organizers implemented stationary format using a smart trainer (Kickr v4, Wahoo Fitness, Atlanta, USA), which was shipped to all teams aiming to ensure standardized conditions. In this global edition, nine international pilots (eight using surface stimulation and one using an implanted system) were conducting a stationary cycling task with minimal resistance, targeting a virtual distance of 1200 m within eight minutes. According to the official results [5], five pilots successfully completed the 1200 m, achieving average speeds between 18.4 km/h and 27.0 km/h. These substantially higher velocities compared to the 2016 over-ground may indicate potential limitations in the use of smart trainers for assessing FES-Cycling performance.

In the third edition of the Cybathlon in 2024, the organizers opted for a virtual race scenario using a stationary smart trainer in combination with the IndieVelo platform (TrainingPeaks, Louisville, USA). Pilots were required to complete a distance of 1960 m within 8 min on a flat virtual track. The aim was to promote an inclusive approach that enabled both in-person and remote participation, while enhancing the engagement of stationary FES-Cycling and allowing for the collection of detailed performance data to support more targeted training strategies. However, commercially available devices are generally not designed to accurately measure the low power outputs typically produced during FES-Cycling (e.g. steady-state ~ 16 W and peak ~ 40 W) [6]. Since the smart trainer used was unable to reliably detect power below 20 W, the organizing committee permitted the use of pedal-based power meters (PM) (PowrLink Zero Dual, Wahoo Fitness, Atlanta, USA), which were able to provide measurement data even below 10 W. However, to date, the accuracy of power measurements in this low-power range has not been clearly established. Moreover, FES-Cycling commonly requires the use of ankle orthoses to stabilize paralyzed lower limbs. Although essential for fixation and safety, these orthoses could potentially alter the transmission of force through the pedals and therefore may affect the recorded power values.

The objective of the present work was to provide a comprehensive summary of the FES-Cycling competition at the Cybathlon 2024. To enable for a more accurate interpretation of the obtained results, we evaluated the measurement accuracy of pedal-based power meters, especially for the low-power-range (< 25 W). Further, we investigated the influence of different external factors such as: mounting torque and orthosis position on the accuracy of the power measurement. We hypothesized that pedal-based power measurements would exhibit systematic deviations in the low-power range and would be sensitive to different mounting conditions. A crank-based power meter served as reference for evaluating the accuracy of the pedal-based system. Both measurement systems were integrated into the same stationary cycling setup to ensure controlled test conditions.

Materials and methods

FES-Cycling at the Cybathlon 2024

A total of eight teams from seven countries, comprising ten pilots, participated in the FES-Cycling discipline of the Cybathlon 2024. The competition took place over two days. On the first day, each pilot completed one qualification run, while on the second day the four fastest pilots advanced to the final race. Seven teams competed on-site at the main hub in Zurich, while one team with two pilots (BeAgain, South Korea) participated remotely from their local hub. All pilots successfully completed their runs without technical failures of their cycling devices.

Pilot eligibility and medical screening

Pilots were only allowed to participate if they had a spinal cord injury with paraplegia and a complete loss of voluntary motor function in the lower limbs (AIS A or B). However, it was required that the pilots retained sufficient voluntary control of the trunk, arms, and neck to safely operate the cycling device. Eligibility for pilots with partial impairment of upper-body control was determined individually. All candidates underwent a standardized medical screening procedure (“MedCheck”), conducted via an online meeting by a neutral medical examiner appointed by the Cybathlon organizing committee. A practicing neurologist, selected by the pilot’s team, was required to attend alongside the pilot. The MedCheck verified compliance with general and discipline-specific eligibility criteria and ensured medical safety for participation. Subject specific characteristics are summarized in Table 1.

Table 1.

Summary of demographic data of all participating pilots based on registration information provided for the Cybathlon 2024

Team	Country	Gender	Lesion level	AIS score	Lesion Since	Pilot weight [kg]	Age [years]	Height [cm]
BeAGain	South Korea	male	T10	A	11/2021	78	52	181
BeAGain 2	South Korea	female	T4-T5	A	06/2022	46	20	160
BFH-CybaTrike	Switzerland	male	C7-C8	A	2012	74	45	184
EMA	Brazil	male	T9	A	2012	78	42	170
FreeWheels BR	Brazil	male	T6	A	2007	72.5	42	176
HunFess	Hungary	male	T7	A	2012	96	55	179
Polimi 2	Italy	male	T5	A	2019	76.1	43	175
PULSE Racing	Netherlands	male	T4-T5	B	06/2020	80	33	179
Vienna FES 1	Austria	male	T3	B	09/2021	93.3	40	177
Vienna FES 2	Austria	male	T7	A	08/2020	63.2	37	171

Cycling devices

All cycling devices were required to comply with the discipline-specific technical regulations to ensure pilot´s safety. Both commercially available systems and custom-built prototypes were permitted, provided that actuation conducted exclusively through the pilot’s legs. Any stimulator capable of activating lower-limb muscles or stimulating spinal neural structures was allowed, as long as it met standard electrical safety regulations (IEC 60601-1 and 60601-2-10 or national equivalent). Devices could employ any number of channels for stimulating lower-extremity muscles.

Although the competition took place on a stationary smart trainer, each device was required to demonstrate full functionality for untethered over-ground cycling. Pilots were allowed to adjust stimulation intensity and patterns during the race, using either manual triggering or closed-loop control strategies. However, an immediate emergency-stop capability had to be ensured at all times.

Compliance with these requirements was confirmed through a standardized technical inspection procedure (“TecCheck”) conducted by the organizing committee. Device specific characteristics and stimulation strategies for each pilot are summarized as supplementary table S1.

Race description

The FES-Cycling discipline was carried out as a stationary race (see Fig. 1) within the indieVelo (v0.3.15, TrainingPeaks, Louisville, USA) virtual race environment, with up to four pilots competing at the same time. All pilots used a self-calibrating smart trainer (KICKR V6, Wahoo Fitness, Atlanta, USA) controlled by the indieVelo application, which rendered the virtual race environment and recorded cycling performance. Each smart trainer was pre-validated by the Cybathlon organizing committee to ensure for accurate power measurements (not disclosed) within ± 1%. As the smart trainer was not designed for low power outputs, it was not able to provide reliable power measurements below 20 W. Thus, a dual-sided pedal based power-meter (PowrLink Zero Dual, Wahoo Fitness, Atlanta, USA) was supplied by the organizers. Each team was allowed to choose between the smart trainer and the pedal power-meter, as primary power-source for the virtual race environment.

Fig. 1.

Overview of the participating teams during the FES-Cycling race at the Cybathlon 2024

The virtual course “Forest Velodrome”, was a flat circular track with an overall distance of 1960 m, to be covered within a time-limit of 8 min. It was divided into ten sections (nine sections of 200 m and a final section of 160 m), each scoring ten points. For each pilot, body weight and height were incorporated into the virtual model, while the weight of the cycling device was not considered.

Each race consisted of a 30 s warm-up followed by an automatic race start. During the first 20 s of the warm-up, pilots were allowed to move their legs either by arm propulsion, by FES or with external assistance. However, during the final 10 s before the race start, only FES-evoked leg movement was permitted. During the race, the usage of hands or arms were allowed to be used to push the legs to overcome pedaling dead points sporadically, but not to support on-going propulsion.

All data was recorded by the indieVelo application at a sampling frequency of 1 Hz and stored as “.fit”-files. In addition, the team ViennaFES collected power data using a separate crank-based power-meter (2INPOWER, ROTOR Bike Components, Madrid, Spain) at a system frequency of 200 Hz calculating power for each crank revolution.

Pedal mounting and calibration

Due to the heterogeneous leg-supports used across teams, each group was responsible for designing its own interface to mount the pedal sensors. Based on preliminary evaluations by the organizing committee, orthosis weight was limited to a maximum of 1 kg per leg. Pedals were installed on the tricycle according to the manufacturer’s specifications, using a mandatory 1 mm washer and a tightening torque of 30 Nm, followed by three forceful sprint starts performed by a healthy subject on a smart trainer to obtain stable power readings. In the Wahoo application, auto-calibration was disabled and the physical crank-arm length was entered. To establish a correct zero offset, a manual calibration was performed with the crank arm positioned vertically and the attached leg support hanging freely.

Compensation strategy for power pedals

As a consequence of mounting the required leg-support onto the power pedals, it became apparent that the power pedals provided a considerable over-estimation of the real power produced, despite careful calibration and correct configuration of the physical crank length. To compensate for this, the organizational committee implemented a compensation strategy, in which they calculated an adjusted crank length for each team based on measurements with the Kickr smart trainer as reference. The aim was to reduce the advantage of the power pedals to 5–10% with respect to the Kickr smart trainer.

Thus, all teams were asked to perform a series of measurement prior to the competition. After conducting the previously described mounting and calibration procedure, a healthy individual cycled for 2 min at 40 W (as displayed by the Kickr smart trainer) using the “Forest Velodrome” track in indieVelo application to record data. Two independent recordings have been performed for each crank length investigated. The crank length parameter of the power pedals, set via the Wahoo app, was based on each team’s physical crank length and varied by − 20 mm, − 10 mm, − 5 mm, 0 mm and + 20 mm. Figure 2 illustrates example measurements for an original crank length of 170 mm.

Fig. 2.

Example of compensation measurements for an original crank length of x = 170 mm. The relative error of the power pedals has been calculated in reference to the Kickr smart trainer while cycling at 40 W for 2 min in the IndieVelo virtual race environment. Two measurements have been performed for the configured crank lengths: x – 20 mm, x – 10 mm, x – 5 mm, x, x + 20 mm, where x denotes the real physical crank length

The day before the qualification, the organization committee conducted additional verification measurements on-site (i.e. 2 independent recordings of 2 min cycling at 40 W on the “Forest Velodrome” track in indieVelo), before announcing the individually adjusted crank length to be used by each team.

For the Hungarian HunFESS team, the calculated crank-length was below the minimal settable crank-length in the Wahoo application. Thus, the Cybathlon committee decided to allow them to participate, but reevaluated their results in post-processing based on a compensation model (see Supplementary material S01_Cybathlon2024_FES_HunFES_correction.pdf) developed by the committee’s experts.

Data analysis

All data recorded by the IndieVelo application were stored as “.fit” files and subsequently converted to “.csv” format using the official FitSDK software package (release 21.141.00, Garmin Ltd., Olathe, USA). Data analysis was performed in Matlab (R2017b, MathWorks, Natick, USA) on data from the final race runs of the finalists and the qualification runs of all other pilots. Race onset was defined as the first data point with a speed greater than 0 km/h, while race termination was set at either completion of the full race distance of 1960 m or expiration of the 8-min time limit. For each race, total distance covered, race duration, maximum speed, mean speed, peak power and mean power were calculated.

Power recordings from the Kickr smart trainer were observed to be stable above a threshold of 20 W. To assess potential inaccuracies of the pedal-based power meters during the race, only data segments in which Kickr power exceeded this threshold continuously for at least 3 s were considered valid. For each race, mean power was calculated across all valid data points separately for the pedal power meters and the Kickr smart trainer. For the ViennaFES team, the same analysis was additionally performed using data from their crank-based power meter.

Accuracy assessment of power pedals

In addition to the competition, we conducted a detailed assessment of the measurement accuracy of the power pedals using a crank-based power meter (2INPOWER, ROTOR Bike Components, Madrid, Spain as reference. Both systems were calibrated according to the manufacturers’ instructions, setting the correct physical crank length of 170 mm on each device. Power data from both meters were recorded concurrently using a Garmin sports watch (Fenix 7, Garmin Ltd., Olathe, USA) through an additional data field [6]. Data were exported from Garmin Connect as .fit files at a sample rate of 1 Hz.

All measurements were performed on a recumbent trike (VTX, ICE Trikes, Cornwall, UK) mounted on a conventional home trainer (In’Ride 100, Van Rysel, Lille, France). The pedal manufacturer specifies the use of a torque wrench with a mounting torque of 30 Nm to ensure secure attachment of the pedals to the crank and accurate measurements. Because this requirement is frequently overlooked in practice, we investigated the influence of pedal mounting torque by repeating measurements at 10, 20, 30 and 40 Nm.

Power pedals are typically designed for use with dedicated cycling shoes (i.e., shoes that attach to the pedals via cleats), which provide a consistent point of force application used in the internal power calculation. In FES-Cycling, however, the paralysed lower limb must be stabilized using an ankle orthosis, which may alter load transfer and lever arms. To evaluate the impact of orthosis use and positioning, measurements with the original pedal configuration (POS ORIG) were compared to measurements obtained with an ankle orthosis (ergopedals with calf support, HP-Velotechnik GmbH, Kriftel, Germany). The orthosis was installed onto the power pedals via a custom adapter plate (see Fig. 3), enabling two leg positions: a narrow configuration (POS 0) and a wide configuration (POS 1). Relative to POS ORIG, the foot midline was shifted laterally by + 6 cm (POS 0) and + 12 cm (POS 1).

Fig. 3.

Adapter plate and mounting positions. Images depicting the custom-build adapter plate at different mounting configurations used during the measurements. The ankle orthosis was mounted to the power pedals via a custom-build adapter plate, which allowed for fixation at different positions. Using POS ORIG as the reference configuration, the midline of the foot was shifted laterally by 6 cm (POS 0) and 12 cm (POS 1)

Measurement protocol

A healthy, able-bodied participant (38 years; 85 kg; 195 cm) performed cycling without FES at target cadences of 25, 30, 40, 50, 60, 70, 80, 90 and 100 rpm. Each cadence level was sustained for 30 s and paced using a metronome. To reflect the low power output typically observed in FES-Cycling, a low gear ratio was selected (26T chain-ring front, 51T rear sprocket) with the home trainer set to its minimum resistance.

Data analysis

Recorded .fit files were converted to CSV using the official FitSDK (release 21.141.00, Garmin Ltd., Olathe, USA). Data processing and analysis were performed in Matlab (R2017b, MathWorks, Natick, USA).

For each of the power meters, all recorded power samples were sorted by the corresponding cadence to derive a model (see Eq. 1) to predicting power (p) from cadence (c). A third-order polynomial with k₀ fixed at 0 W was fitted by estimating k₁, k₂ and k₃ to minimize the overall fitting error. Fixing k₀ = 0 W constrains the model to ensure the condition of zero power at zero cadence, thereby preventing bias from extrapolation in the low-cadence range. For device comparison, the mean model-predicted power was computed for both systems over the cadence range 0–100 rpm.

1

The variability around the fitted model was quantified using the root-mean-square error (RMSE) across all measured samples. RMSE (see Eq. 2) was calculated separately for each power meter using its own fitted model.

2

The mean model-predicted power over the cadence range of 0–100 rpm was computed for both power meters and used to derive a compensated crank length L_comp according to Eq. 3, using a native reference crank length L_ref = 170 mm.

3

To assess the feasibility of compensating potential deviations of the pedal power meter, a second independent measurement was conducted using the compensated crank length rounded to the nearest 0.5 mm.

Results

FES-Cycling at the Cybathlon 2024

Data were available for eight of the ten participating pilots, as one team (BeAGain) did not provide their original files. Figure 4 illustrates the progression of covered distance over time for each pilot.

Fig. 4.

Distance covered over time during the FES-Cycling race at the Cybathlon 2024. The figure depicts the temporal progression of the distance covered for each analyzed team. Races terminated either upon completion of the total race distance of 1960 m (black line) or upon reaching the 8-min time limit (blue vertical line). The total race distance of 1960 m was divided into 10 segments (horizontal dotted lines), used for scoring. The last segment completed has been marked (black circles) for each team

Figure 5 presents an overlay of the power recordings, comparing pedal-based power meter with the respective Kickr smart trainer of each team. For the ViennaFES team, power data from an additional crank-based system was included. Race onset and termination, as well as data segments classified as valid based on a Kickr power threshold > 20 W, are indicated. Across all analyzed teams using pedal-based power meters as the primary recording device, pronounced discrepancies between pedal-derived and Kickr-derived power values were observed.

Fig. 5.

Comparison of output power during the FES-Cycling race at the Cybathlon 2024. This figure gives a direct comparison between the pedal based power meter (red traces) and the Kickr smart trainer (black traces) across all analyzed teams. For the ViennaFES team, power data from an additional crank-based power meter are included (green traces). Vertical lines indicate race start (green), segments (dotted black), last segment completed (red) and the 8-min time limit (blue). The 30 s preceding the start correspond to a defined warm-up period, during which initial leg motion could be assisted before transitioning to FES-only activation to enable a rolling start. Shaded regions highlight areas in which Kickr power exceeded the defined threshold of 20 W

Table 2 summarizes the race outcomes derived from the .fit-file analysis. Teams were ranked according to the analyzed performance metrics, excluding post-race penalties applied to the HUNFESS team, while the official Cybathlon ranking is reported for comparison. The additional ranking is included solely to illustrate the influence of measurement uncertainties on performance metrics and does not aim to reinterpret or challenge the official Cybathlon results. Results for the BeAGain team were taken from self-reported data provided by the team.

Table 2.

Results of the FES-Cycling race at the Cybathlon 2024

Ranking by data	Ranking by Cybathlon	Team	Country	Power meter	Distance of the last segment completed	Time of the last segment completed	Time of the last segment completed	Speed max	Speed mean	Peak power	Power mean	n valid (P Kickr > 20 W)	Mean Power Kickr valid (REF)	Mean power pedals valid	Relative error pedals	Mean power crank valid	Relative error crank
					m	s	mm: ss	km/h	W	W	W	#	W	W	%	W	%
1	1	BeAGain	South Korea	Pedal PM	1960	362	06:02	22.9*	19.5*	111*	56.2*	-	-	-	-	-	-
2	2	PULSE Racing	Netherlands	Smart trainer	1960	364	06:04	21.5	19.3	88	60.9	365	60.9	-	-	-	-
3	3	Polimi 2	Italy	Pedal PM	1960	456	07:36	20.0	15.5	68	35.2	262	35.4	43.8	23.8	-	-
4	4	Vienna FES 1	Austria	Pedal PM	1960	479	07:59	17.3	14.7	54	37.2	268	24.6	40.9	66.6	28.3	15.2
5	5	BFH-CybaTrike	Switzerland	Pedal PM	1400	438	07:18	15.6	11.5	58	20.4	10	33.0	47.8	44.8	-	-
6	6	EMA	Brasilia	Pedal PM	1400	464	07:44	12.9	10.9	44	19.5	26	26.5	30.3	14.7	-	-
7	8	HunFess	Hungary	Pedal PM	1200	448	07:28	14.9	9.6	49	18.4	17	24.5	38.4	56.6	-	-
8	7	Vienna FES 2	Austria	Pedal PM	1000	360	06:00	15.5	10.0	38	16.0	96	23.1	30.9	33.8	26.2	13.5
9	9	FreeWheels BR	Brasilia	Pedal PM	600	331	05:31	10.2	7.8	22	11.0	0	-	-	-	-	-
10	10	BeAGain 2	South Korea	Pedal PM	400	123	02:03	22.3*	11.7*	66*	4.0*	-	-	-	-	-	-

Team performance metrics were derived from analysis of the .fit files recorded during the race via the indieVelo application

In line with the Cybathlon rules, distance and time were provided from the last segment completed

The data-driven ranking excluded post-race penalties for the HUNFESS team, resulting in differences from the official Cybathlon ranking

To assess the accuracy of the pedal-based power meters, mean power was calculated across all n valid data points (Kickr power > 20 W) and compared with the Kickr smart trainer

Where available, crank-based power meter data are additionally reported. Data marked with an asterisk (*) were self-reported by the team, raw data not available

Consistent with the observations in Fig. 5, a systematic overestimation of power by the pedal-based systems was observed, ranging from 14.7% (Δ = 3.9 W, n = 26 data-points) to 66.6% (Δ = 16.4 W, n = 268 data-points) relative to the Kickr smart trainer within valid segments, where n denotes the total number of 1-Hz data points aggregated across all valid segments for the respective race. For the FreeWheels BR team, no valid segments could be identified, as the produced power remained below the detection limit of the Kickr smart trainer. For the ViennaFES team, the crank-based power meter also reported higher values than the Kickr, with deviations ranging from + 13.5% (Δ = +3.1 W, n = 96) to + 15.2% (Δ = +3.7 W, n = 268) within valid segments

Accuracy assessment of power pedals

Power data from pedal-based meters were compared with a crank-based reference under controlled conditions. Representative raw data and the corresponding cadence-based power prediction models for both devices are shown in Fig. 6.

Fig. 6.

Measurement data and power prediction model. Representative example (POS 0, mounting torque 30 Nm) comparing pedal-based power measurements (red traces and data points) against the crank-based power meter used as reference (black traces and data points). A Raw data of recorded power and cadence. B Power data sorted by cadence with the corresponding fitted power prediction models

The influence of pedal mounting torque and ankle orthosis mounting position is illustrated in Fig. 7 by comparing the power prediction models of both measurement systems. Variations in mounting torque resulted in slight changes in the power data recorded by the pedal-based system, whereas the crank-based reference remained largely unaffected. Introduction of the ankle orthosis and lateral displacement of the foot further altered pedal-derived power relative to the crank reference. Figure 7 further demonstrates that adjusting the crank length effectively reduced deviations between pedal- and crank-based power models, as confirmed by a second independent measurement for each condition.

Fig. 7.

Influence of mounting torque, mounting position and crank-length compensation. Comparison of power prediction models obtained from the pedal-based power meter (red traces) and the crank-based reference (black traces) under different conditions. The effect of crank-length compensation was evaluated using a second independent measurement for each condition. A Variation of pedal mounting torque (10, 20, 30, and 40 Nm). B Effect of ankle orthosis mounting position, introducing a lateral shift of the foot midline by + 6 cm (POS 0) and + 12 cm (POS 1) relative to the original configuration without orthosis (POS ORIG)

Quantitative results of the accuracy assessment are summarized in Table 3. For each condition, relative error, RMSE (see Eq. 2) and compensated crank length were derived from the recorded data using the model-based analysis. Across all investigated conditions, use of the native crank length of 170 mm resulted in deviations in recorded power relative to the crank-based reference. The application of a compensated crank length reduced the relative error in all conditions within the investigated low-power range (< 25 W).

Table 3.

Results of the accuracy assessment under different conditions

Test condition	Native (Crank length 170 mm)					Compensated
	Mean power CRANK (Ref)	Mean Power PEDALS	Relative error	RMSE CRANK (Ref)	RMSE PEDALS	Compensated crank length	Mean power CRANK (Ref)	Mean power PEDALS	Relative error
	W	W	%	W	W	mm	W	W	%
Mounting torque 10 Nm	12.0	12.7	6.2	0.5	1.2	161.0	11.8	12.0	1.8
Mounting torque 20 Nm	11.5	11.0	-4.1	0.4	1.1	178.0	11.3	11.3	-0.7
Mounting torque 30 Nm	11.5	10.9	-5.2	0.4	1.1	180.0	11.4	11.5	1.5
Mounting torque 40 Nm	11.5	10.2	-11.9	0.4	1.6	193.5	11.8	11.7	-0.4
POS ORIG	12.5	12.9	2.7	0.3	0.8	166.5	11.8	11.9	0.9
POS 0	13.0	15.6	20.1	0.4	2.6	142.0	13.2	12.7	-3.4
POS 1	11.3	14.6	29.8	0.5	3.5	131.5	12.0	11.3	-6.4

Relative error between the pedal-based power meter and the crank-based reference was calculated from mean recorded power values

Raw data were further used to derive power prediction models (see Eq. 1)

Calculate the root-mean-square error (RMSE) as defined in Eq. 2, and determine a compensated crank length (see Eq. 3)

The effectiveness of the compensation was assessed in a second independent measurement for each condition

Figure 8 provides a comparison of relative error and RMSE (see Eq. 2) across all conditions. Relative error was consistently reduced when applying the compensated crank length compared with the native reference length. RMSE values indicated higher variability of the pedal-based system across measurement conditions, whereas the crank-based system remained largely unaffected.

Fig. 8.

Comparison of relative error and root-mean-square error (RMSE) under different conditions. (A) Relative error between the pedal-based power meter and the crank-based reference was calculated from mean recorded power values. Relative error between native (L_ref = 170 mm) and compensated crank length obtained from independent measurements for each condition. (B) RMSE (see Eq. 2) was calculated across all samples for each fitted model, separately for each power meter. The figure compares the pedal-based power meter to the crank-based reference at the native crank length (Lref = 170 mm)

Discussion

FES-Cycling at the Cybathlon 2024

As in previous editions, the Cybathlon served as a highly motivating platform for the development and evaluation of assistive technologies, fostering exchange between researchers, engineers, and pilots. In 2024, the FES-Cycling discipline included 10 pilots from 8 international teams, making it the third-largest discipline after the arm prosthesis race (ARM) and leg prosthesis race (LEG) [7]. While most teams competed on site in Zurich, the South Korean team BeAGain participated remotely via a local hub, further highlighting the flexibility of the competition format.

At the Cybathlon 2024, the FES-Cycling race was conducted on a Kickr smart trainer within the virtual race environment indieVelo, with the aim of increasing pilot engagement and standardizing race conditions. Virtual race environments offer the advantage of enabling competition with remotely connected participants, a feature that has gained particular relevance in the context of the COVID-19 pandemic. Prior work has suggested that virtual cycling scenarios can be used to assess physiological performance using commercially available devices in able-bodied athletes [8]. However, it remains unclear whether these devices are able to provide reliable data for the low-power outputs (< 25 W) typically generated during FES-Cycling by individuals with SCI [6].

For the competition, the organizational committee validated the accuracy of each Kickr smart trainer prior to distribution. In collaboration with engineers from Maxon Motors, the trainers were tested in a virtual cycling scenario using motor-driven constant cadence. Across all devices, the time required to complete a 2000 m course with slopes of up to ± 2% deviated by less than ± 1%, corresponding to an average power output of approximately 37 W (unpublished data). However, as the Kickr smart trainer was unable to provide stable power measurements below 20 W, a dual-sided pedal-based power meter (PowrLink Zero Dual) was introduced to enable power detection even below 10 W. While this limitation of the Kickr is unlikely to reflect a fundamental constraint of drivetrain-based power measurement, addressing it would have required a dedicated firmware validated for the low-power range, which was not available from the manufacturer at the time.

For the pedal-based system, the manufacturer-stated accuracy of ± 1% was accepted by the committee without further validation. Following feedback from participating teams, a systematic overestimation of pedal-derived power became apparent. In response, the organizing committee introduced a standardized mounting and calibration protocol and assigned an individual virtual crank length to each team. This adjusted crank length was based on pre-race measurements, aiming to limit the advantage of the pedal-based system relative to the smart trainer to within 5–10%.

During the competition, four pilots completed the full virtual race distance of 1960 m within the 8-minute time limit. The remaining pilots covered distances between 400 m and 1400 m, with respect to their last complete segment. Mean cycling speeds ranged from 7.8 to 19.5 km/h, with maximum speeds between 10.2 and 22.9 km/h. Compared with the over-ground Cybathlon 2016 event [4], where surface-stimulated pilots achieved average speeds of 5.7–11.2 km/h, the virtual race environment resulted in substantially higher velocities. This difference is likely attributable to the virtual model considering only pilot body weight, while neglecting the mass of the cycling device. This highlights a fundamental limitation of virtual FES-Cycling competitions: while they ensure standardized conditions, they are not directly comparable to over-ground racing and likely overestimate achievable real-world velocities. This effect becomes particularly pronounced at low power outputs, where small absolute differences can result in large relative changes in virtual speed.

With respect to power output, the Dutch team PulseRacing deserves special attention, as it was the only team using the validated Kickr smart trainer as the primary power source during the race. The pilot achieved a mean power output of approximately 61 W, with peak values up to 88 W, markedly exceeding typical power levels reported for FES-Cycling. For example, Duffell et al. reported steady-state powers of 16.0 ± 3.6 W and peak powers of 40.8 ± 4.6 W in individuals with SCI [6], consistent with other published studies [9–11]. Pedal-based mean power outputs ranged from approximately 11 W to 56 W, with peak values up to 111 W. However, these values require cautious interpretation. Analysis of valid race segments (Kickr power > 20 W) revealed substantial overestimation by the pedal-based systems, with relative errors of 14.7–66.6%, far exceeding the intended compensation target of 5–10%. Additional crank-based measurements provided by the ViennaFES team indicated mean deviations of + 3.1 to + 3.7 W relative to the Kickr smart trainer. While this deviation is most likely attributable to drivetrain transmission losses, it is important to acknowledge that, at such low absolute power outputs, even small differences may translate into a meaningful advantage in the overall ranking.

An additional limitation concerns resistance control of the Kickr smart trainer at low power outputs. As the device does not reliably measure power below 20 W, precise resistance regulation in this range appears unlikely. Pilots frequently reported an unnatural, intermittent “kicking” sensation during cycling, suggesting discrete resistance switching rather than continuous control. Beyond discomfort, this behavior complicates stimulation control when aiming to achieve smooth pedaling during FES-Cycling.

Accuracy assessment of power pedals

An additional objective was to quantify the accuracy of the pedal-based power meter against a crank-based reference for the low-power range (< 25 W), with specific focus on the influence of pedal mounting torque and ankle-orthosis mounting position.

Mounting torque affected pedal-derived power, most notably at the highest torque: at 40 Nm, the pedals underestimated power by 11.9% relative to the crank-based reference and exhibited a higher RMSE of the fitted model (1.6 W vs. 0.3 W). For the remaining torques, deviations ranged from − 5.2% to + 6.2% with RMSE values of 1.1–1.2 W. As shown in Fig. 7, applying a compensated crank length markedly reduced these deviations, yielding residual errors of − 0.4% to + 1.8% across all tested torques.

Orthosis mounting position had a larger impact on measurement accuracy. In the original configuration (POS ORIG), relative deviation was small (2.8%) with an RMSE of 0.8 W, whereas orthosis mounting led to substantial overestimation (POS 0: +20.1%; POS 1: +29.8%) and increased RMSE, most pronounced at POS 1 (3.5 W). Compensation reduced the relative error to − 3.4% (POS 0) and − 6.4% (POS 1). Nevertheless, Fig. 7 shows a drop in the pedal-derived power model at higher cadences, suggesting that shifts in the effective point of force application can compromise the internal force model of the power meter. Considering that pedal-based power meters are optimized for rigid cycling shoes, they rely on predictable force vectors. In FES-cycling orthoses are required to stabilize the paralyzed leg, which may alter lever arms and load transfer. Although the crank-based reference was less affected, it also exhibited a slight positive bias. Based on these data, the orthosis should be mounted as close to the crank arm as feasible to minimize lateral displacement of the foot midline.

A limitation of this work was that all measurements were performed by a single healthy participant to ensure a controlled test setting. Nevertheless, interpreting these findings as representative examples of potential sources of bias may contribute to a better understanding of the deviations observed during the Cybathlon 2024 FES-Cycling competition. Interestingly, preliminary tests using the same pedal system revealed even greater power overestimation [12], underscoring the substantial variability of pedal-based power measurements in the low-power range.

In the competitive context, an additional consideration arises from the design of the dual-sided pedal power meter. Although power is computed independently by each pedal, only the left pedal acts as the master unit and transmits data from both pedals. If the wireless connection to the right pedal is lost or the pedal is unpowered, only unilateral power is recorded, resulting in half of the total output. While such failures are typically detected quickly in elite cycling, they may remain unnoticed in FES-Cycling, where absolute power output is low and generally unknown.

In summary, our findings indicate that currently neither pedal-based power meters nor the Kickr smart trainer are well suited for establishing a low-power FES-Cycling competition within a virtual race environment.

From a competition-design perspective, the virtual race environment represented a pragmatic compromise between standardization, accessibility and logistical feasibility, despite its technical limitations at low power outputs. From a competition-design perspective, the virtual race environment represents a pragmatic compromise between standardization, accessibility and logistical feasibility. It may further serve as a valuable complementary approach for enabling engaging training and more frequent competition opportunities, provided that reliable measurement systems for the low-power range become available. Although continuous power data throughout the race are of high scientific interest, the presented results support renewed consideration of over-ground race formats, in which power measurement plays a less central role in competitive ranking.

Virtual race environments may represent a valuable complementary approach, enabling accessible training and more frequent competition opportunities, provided that reliable measurement systems for the low-power range become available.

Over-ground competitions are more closely aligned with the overarching goal of improving mobility in individuals with spinal cord injury and are more effective in motivating pilots and engaging the general public. Recent developments, including assistive-technology record races (ETH-Zurich) [13], the Lyon Cyber Days (ANTS, ENS Lyon) [14] and the ViennaFES Cybersportday 2025 (Medical University of Vienna) [15], reflect a clear direction for the future of FES-Cycling competitions.

Conclusions

FES-Cycling competitions, such as those conducted at the Cybathlon 2024, provide an important framework for evaluating assistive technologies under standardized conditions. Through a virtual race environment the organizers aimed for global accessibility and while providing detailed performance data. However, this work demonstrates that current commercially available pedal-based power measurement systems and smart trainers are not yet suitable for ensuring a reliable virtual competition at the low power outputs typical of FES-Cycling. In line with the overarching goal of improving mobility for individuals with SCI, future FES-Cycling competitions should prioritize over-ground race formats, which assess externally observable outcomes such as distance and time. While power measurements remain of high scientific interest, they should not be used as only metric to assess cycling performance in a competitive setting. Over-ground races further tend to enhance pilot motivation and public engagement. Emerging competition formats, including assistive-technology record races (ETH Zurich) and community-based events such as the Lyon Cyberdays (ANTS, ENS Lyon) and the ViennaFES Cybersportday 2025 (Medical University of Vienna), point toward a clear and promising direction for the future of FES-Cycling competitions.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

A

Stimulation amplitude

AIS

American spinal injury association impairment scale

F

Stimulation frequency

FES

Functional electrical stimulation

G

Gluteal muscles

H

Hamstrings muscles

PhW

Phasewidth

PM

Power meter

Q

Quadriceps muscle

RMSE

Root-mean-square error

SCI

Spinal cord injury

TS

Triceps surae

.csv

Comma separated values – text based file format

.fit

Flexible and Interoperable Data Transfer - proprietary file format used by Garmin

Author contributions

All authors contributed to the study conception and design. MS, QN, BR and EK were responsible for building the technical test setup. Material preparation, data collection were performed by MS, RSDB, TPAR, EA, NS, EAA, RDSB, HRM, MMDO, JL and AF. Data analysis and presentation was performed by MS. AP and RC provided valuable clinical input and supported the study from a medical perspective. The first draft of the manuscript was written by MS and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. The study was supervised by AP, WM and RC, while MS acquired financial resources for funding this publication.

Funding

No specific external funding was received for this work. Participation in the Cybathlon 2024 was financed individually by the participating teams.

Data availability

The data analysed in this study are included in the manuscript. Raw data are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval was not required for this study, as all data were collected in the context of a public sportive competition. As part of the Cybathlon 2024 registration process, all participating pilots gave their written informed consent to participate in the event while acknowledging potential risks.

Consent for publication

The team managers obtained informed consent from their respective participants for the use of images, as well as personal and measurement data, in this publication.

Competing interests

The authors declare no competing interests.