Towards a Smart Luge that Measures Steering Input of the Rider

Rupert Staudinger

1,2

, Wolfgang Kremser

1,*

, Christoph Thorwartl

and Hermann Schwameder

Salzburg Research Forschungsgesellschaft mbH, Salzburg, Austria

Department of Sports and Exercise Science, Paris-Lodron University Salzburg, Austria

Keywords: Body Sensor Networks, Data-Based Training, Luge, Smart Sports Equipment, Winter Sports.

Abstract: Luge is a high-speed Olympic Winter sliding sport that is timed in milliseconds. The athlete’s steering

performance is a crucial factor for success, but there are currently no objective methods to evaluate steering

technique and timing. As a work in progress, we present a lab prototype of the ‘smart luge’, a sled retrofitted

with six unobtrusive commodity force sensors. The results of a laboratory test with five simulated runs show

that the current setup is capable of measuring the athlete’s activity during steering. This work aims to advance

data-supported training in the luge sport by enabling the in situ measurement of luge athletes’ activity.

1 INTRODUCTION

Luge is an Olympic Winter sliding sport in which a

single or a pair of athletes (‘lugers’) compete for the

shortest time riding a sled down an icy track. Luge is

also the name of the sled that is used.

While descending the track, the luger’s main

influence on their runtime is their steering

performance. The ideal strategy is to stay on the

shortest path downward with minimal steering in

terms of frequency and magnitude (Gong et al.,

2016). Lugers experience speeds over 150km/h

(Schleinitz et al., 2022) so the window for optimal

steering action is extremely small. Even minor

mistakes can cost a race considering that run times are

measured in milliseconds (Platzer et al., 2009).

Trainers currently assess their athlete’s on-track

performance using video analysis (e.g. Fedotova &

Pilipivs, 2010). Given the high speeds and subtle

movements involved in luge steering, this form of

subjective feedback is inherently limited.

In some sports, trainers have already started to

complement their observations with objective data

from sensors that are either worn by the athletes or are

integrated into the sports equipment (Rajšp & Fister,

2020). However, such sensors have not yet been

integrated into luge training, and the scientific

literature on this topic is sparse.

To advance data-driven luge training we started

the development of a ‘smart luge’. The goal is to build

a sensor-equipped luge that can accurately and

Corresponding Author: wolfgang.kremser@salzburgresearch.at

reliably measure steering input. The resulting data

can be analyzed and visualized to give trainers

detailed and objective information on how to improve

their trainee’s steering. This paper presents the first

milestone of the ongoing project, a lab demonstrator

of the ‘smart luge’.

1.1 The Art of Luge Steering

The basic design of a luge consists of a fiberglass

‘pod’ in which the luger lies in a supine position

during the race. The pod is tightly coupled to the left

and right ‘runners’ at the bottom via a steel frame

called the ‘bridge’. The runners are made out of wood

or fiberglass, and they end in upwards ‘bows’ near the

luger’s calves. At the bottom of the runners are the

‘blades’ made from steel that glide on the ice.

In their neutral position, the runners are slightly

bent towards each other. Lugers steer by twisting

them, which causes the blades to cut a leading grove

that gets followed by the luge. The twisting of the

runners can be achieved by a combination of (a)

applying pressure to a bow using the calf, (b) lifting

the bridge using handles that connect the pod and

bridge, and (c) pressing down with one shoulder

(Pareek et al., 2021). Depending on the desired

direction change, these forces are applied differently

between the left-hand and right-hand sides (Figure 1).

206

Staudinger, R., Kremser, W., Thorwartl, C. and Schwameder, H.

Towards a Smart Luge that Measures Steering Input of the Rider.

DOI: 10.5220/0012999400003828

In Proceedings of the 12th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2024), pages 206-210

ISBN: 978-989-758-719-1; ISSN: 2184-3201

Figure 1: Our luge steering model with the anticipated

pressure points drawn in.

1.2 Aim and Research Questions

The ‘smart luge’ aims to unobtrusively measure these

theoretically derived steering movements using

inexpensive commodity sensors in a laboratory

environment. The objectives are (a) to develop a

sensor-based luge prototype, (b) to calibrate the

sensors, and (c) to capture and analyze basic left-right

steering characteristics in a laboratory setting.

2 METHODS

2.1 Instrumentation

According to our theory of luge steering, we should

be able to detect steering maneuvers by measuring the

pressure that (a) the calves apply to the bows, (b) the

hands apply to the handles, and (c) the shoulders

apply to the pod.

We placed a FlexiForce™ sensor (Tekscan Inc.,

USA) at the anticipated steering points shown in

Figure 1. These sensors are 0.2 mm thin force sensing

resistors (FSR) that increase their electrical

conductance in proportion to the force that is acting

on them. We used the largest FSR model (A502) for

the shoulders, the mid-size model (A401) for the

bows, and the smallest model (A301) for the handles.

We used thin double-sided adhesive tape to attach the

sensors to the bows and the pod. The sensors for the

handles were placed at the interface between the

handles and the bridge.

For data acquisition, we used the KRYPTON®

CPU with two strain gauge modules and the

DewesoftX software (Dewesoft, Slovenia). This

setup recorded the FSR sensors’ voltage outputs at

20kHz. The changes in electrical resistance induced

by the FSR sensors were converted into a reciprocal

proportional output voltage (u

). Calibration of the u

was achieved through a 2-point calibration method

using standardized weights, ensuring precise force

measurements.

2.2 Study Design

The first author of the present paper who had

participated in an Olympic luge competition was the

test luger for this pilot study (sex: male, weight:

85 kg, height: 188 cm). The instrumented luge was

placed on top of a table such that the luger faced the

wall. We projected a pre-recorded point-of-view

video of a luge run onto that wall. The track in the

video was familiar to the test luger who was asked to

steer as he would if he had been in the video. The

same run was repeated five times. A webcam that was

synchronized with the sensor hardware recorded the

entire study setup.

We noted the frames in which the luger in the

video entered and exited a curve, as well as the

curve’s direction (left, right), and noted them in an

Excel sheet. We excluded the first curve because it

follows the startup phase where the luger is trying to

gain momentum with their hands in a sitting position.

Thus, we did not consider it a regular curve.

Furthermore, we excluded curve 11 (the ‘Kreisel’)

which requires more complex steering motions and

thus would not be comparable to the other curves of

the track.

2.3 Data Analysis

The resulting data was analyzed with MATLAB (The

MathWorks, Inc., USA). For each run, using the

synchronized webcam footage, we discarded all data

that was recorded before the pre-recorded video

started and after it ended. Then we used the curve start

and end points to segment the remaining data. We

normalized the data for each curve to 1000 samples.

Furthermore, each curve was split into three phases:

‘entry’ (0% - 25% of samples), ‘core’ (25%-75%),

and ‘exit’ (75% - 100%). We plotted the average

signal of each of the six FSR sensors, along with the

standard error, across all five runs.

To quantify the (dis)similarity of the sensor

signals we calculated Pearson’s correlation

coefficient (r) for each pair of sensors and each curve

phase’s mean. Coefficients higher than 0.1, 0.3, 0.5,

and 0.7 represent small, moderate, large, and very

large correlations, respectively (Hopkins et al., 2009).

Towards a Smart Luge that Measures Steering Input of the Rider

207

3 RESULTS

Figure 2 shows the average signal (+/- standard error)

for each sensor with the individual curves colored in.

The maximum values for the bows (left: 12N ± 1,

right: 17N ± 1) and the shoulders (left: 23N ± 3, right:

18N ± 2) differ considerably from the maximum

values of the handles (left: 631N ± 81, right: 404 ±

72).

Figure 3 shows the three correlation matrices, one

for each curve phase. In all three phases, the handles

have a very large correlation (r between 0.85 and

0.87). The left and right shoulders have a consistently

negative correlation (r at entry: -0.27, core: -0.88,

exit: -0.42). The left and right bows show a moderate

positive correlation at curve entry (r=0.37) which

Figure 2: Plots of the calibrated force signal [N] with the standard error of all six FSR sensors averaged over all five runs.

Grey sections mark right curves, orange sections mark left curves. The first curve and the dark gray "Kreisel" section were

excluded from the analysis.

Figure 3: Correlation matrices of the average force signals from the six FSR sensors, grouped by curve phases. Significant

correlations are printed in bold (significance levels *: p < 0.05, **: p < 0.01).

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

208

changes to a negative correlation in the core phase

(r=-0.63) and at the exit (r=-0.35). Furthermore, we

observed that in the core and exit phases, the shoulder

and bow on opposite sides correlate strongly (r

between 0.68 and 0.75) while the shoulder and bow

on the same side have a large negative correlation (r

between -0.81 and -0.52). In general, the mean

absolute r value was highest in the core phase (0.56),

followed by the exit (0.47) and entry (0.41) phases.

4 DISCUSSION

We demonstrated a lab prototype of the ‘smart luge’,

a luge sled that was retrofitted with six FSR sensors

to measure the force that is applied by the luger to

induce steering.

Figure 4 compares the results with our

expectations based on our luge steering model (Figure

1). We found that sensors that we expected to

correlate positively had a very large positive

correlation, and the sensors that we expected to

negatively correlate had a large negative correlation.

What was unexpected were the high peak force values

of the left and right handles) and their continuously

high correlation between the left-hand and right-hand

side.

Figure 4: Correlations between the FSR sensor values in the

core phase. Blue arrows indicate an expected negative

correlation, and black arrows indicate an expected positive

correlation.

One explanation might be the FSR sensor

placement under the screwed-down handles. Since

both handles are tightly coupled with the bridge,

when one handle is pulled, the handle on the opposite

side moves up as well and squeezes the sensor rather

than twisting away as we had expected. Further

attention is necessary to understand the deformations

of the bridge and how they connect to the athlete’s

steering input.

5 CONCLUSION

In light of this pilot study’s results, we consider the

presented ‘smart luge’ demonstrator as capable of

measuring a luger’s steering maneuvers in a

laboratory environment.

The next step would be to test the system on a real

ice track. However, in its current state, the data

acquisition hardware is too bulky to be safely

transported on the luge. Furthermore, because we

expect a considerable amount of vibration on the ice,

a more sophisticated post-processing/filtering of the

FSR sensor signals is likely necessary to detect the

luger’s steering input. Furthermore, we will optimize

the sensors’ surface sizes and geometries to better

detect the applied forces.

REFERENCES

Fedotova, V., & Pilipivs, V. (2010). Biomechanical

Patterns of Starting Technique during Training and

Competitive Events for Junior Lugers. In C. T. Lim &

J. C. H. Goh (Eds.), 6th World Congress of

Biomechanics (WCB 2010). August 1-6, 2010

Singapore (Vol. 31, pp. 282–285). Springer Berlin

Heidelberg. https://doi.org/10.1007/978-3-642-14515-

5_73

Gong, C., Phillips, C. W. G., Rogers, E., & Turnock, S. R.

(2016). Analysis of Performance Indices for Simulated

Skeleton Descents. Procedia Engineering, 147, 712–

717. https://doi.org/10.1016/j.proeng.2016.06.253

Hopkins, W. G., Marshall, S. W., Batterham, A. M., &

Hanin, J. (2009). Progressive Statistics for Studies in

Sports Medicine and Exercise Science. Medicine &

Science in Sports & Exercise, 41(1), 3–12.

https://doi.org/10.1249/MSS.0b013e31818cb278

Pareek, A., Martin, R. K., & Engebretsen, L. (2021). Luge,

bobsleigh, skeleton. In S. Rocha Piedade, P. Neyret, J.

Espregueira-Mendes, M. Cohen, & M. R. Hutchinson

(Eds.), Specific sports-related injuries (pp. 329–339).

Springer International Publishing. https://doi.org/10.

1007/978-3-030-66321-6_23

Platzer, H.-P., Raschner, C., & Patterson, C. (2009).

Performance-determining physiological factors in the

luge start. Journal of Sports Sciences, 27(3), 221–226.

https://doi.org/10.1080/02640410802400799

Rajšp, A., & Fister, I. (2020). A Systematic Literature

Review of Intelligent Data Analysis Methods for Smart

Towards a Smart Luge that Measures Steering Input of the Rider

209

Sport Training. Applied Sciences, 10(9), 3013.

https://doi.org/10.3390/app10093013

Schleinitz, J. V., Wörle, L., Graf, M., & Schröder, A. (2022).

Modeling ice friction for vehicle dynamics of a bobsled

with application in driver evaluation and driving

simulation. Tribology International, 165, 107344.

https://doi.org/10.1016/j.triboint.2021.107344.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

210