This item was submitted to Loughborough's Research Repository by the author.

Items in Figshare are protected by copyright, with all rights reserved, unless otherwise indicated.

Maximising somersault rotation in tumbling

PLEASE CITE THE PUBLISHED VERSION

http://dx.doi.org/10.1016/j.jbiomech.2003.09.008

PUBLISHER

© Elsevier

VERSION

AM (Accepted Manuscript)

LICENCE

CC BY-NC-ND 4.0

REPOSITORY RECORD

King, Mark A., and Maurice R. Yeadon. 2019. “Maximising Somersault Rotation in Tumbling”. figshare.
https://hdl.handle.net/2134/6753.
 This item was submitted to Loughborough’s Institutional Repository
(https://dspace.lboro.ac.uk/) by the author and is made available under the
following Creative Commons Licence conditions.

For the full text of this licence, please go to:

http://creativecommons.org/licenses/by-nc-nd/2.5/
Journal of Biomechanics 37, 471-477, 2004

Maximising somersault rotation in tumbling

Mark A. King and Maurice R. Yeadon

School of Sport and Exercise Sciences, Loughborough University, Loughborough LE11 3TU, UK.

ABSTRACT

Performing complex somersaulting skills during the flight phase of tumbling requires the
generation of linear and angular momenta during the approach and takeoff phases. This paper
investigates how approach characteristics and takeoff technique affect performance with a view to
maximising somersault rotation in tumbling. A five-segment planar simulation model, customised
to an elite gymnast, was used to produce a simulation which closely matched a recorded
performance of a double layout somersault by the elite gymnast. Three optimisations were carried
out to maximise somersault rotation with different sets of initial conditions. Using the same initial
linear and angular momentum as the double layout somersault and varying the joint torque
activation timings allowed a double straight somersault to be performed with 19% more rotation
potential than the actual performance. Increasing the approach velocity to a realistic maximum of
7 ms-1 resulted in a 42% reduction in rotation potential when the activation timings were
unchanged but allowed a triple layout somersault to be performed with an increase of 31% in
rotation potential when activation timings were re-optimised. Increasing also the initial angular
momentum to a realistic maximum resulted in a 4% reduction in rotation potential when the
activation timings were unchanged but allowed a triple straight somersault to be performed with a
further increase of 9% in rotation potential when activation timings were re-optimised. It is
concluded that the limiting factor to maximising somersault rotation is the ability to generate high
linear and angular velocities during the approach phase coupled with the ability to adopt consonant
activation timings during the takeoff phase.

Keywords : tumbling, gymnastics, simulation model, optimisation

INTRODUCTION

Tumbling is a dynamic activity performed by gymnasts and tumblers from an elastic

takeoff surface. In Artistic Gymnastics, gymnasts perform tumbling sequences on a sprung
floor which has a diagonal length of 17 m whereas in tumbling there is a 26 m sprung track
with an additional 11 m run-up. As a consequence tumblers are able to generate more linear
and angular momenta during the approach phase. A typical tumbling sequence starts with an
approach run where linear momentum is generated, followed by a round-off and flic-flac,
during which angular momentum is produced, and culminates in a somersaulting skill.
During the somersault takeoff phase the gymnast is able to change the linear and angular
momenta by applying muscular torques. The performance of the somersaulting skill is
dependent on the linear and angular momenta at takeoff and the configuration changes used
by the gymnast during flight. The two most important factors for successful performance are
the vertical velocity of the mass centre and the angular momentum about the mass centre at

1
takeoff (Brüggemann, 1983, 1987; Hwang et al., 1990) since the product of these two factors
dictates how much somersault rotation can be achieved.
The characteristics of the approach are obviously important for a successful
performance in tumbling. For example Brüggemann (1987) showed that the angular
momentum and horizontal velocity at touchdown were closely related to the height achieved
in the flight phase (r = 0.81). For the maximisation of somersault rotation it might be
expected that a faster approach will be better, since this will result in more energy at
touchdown and the potential to have more energy at takeoff. However upper limits for the
generation of angular momentum and horizontal approach velocity have not been quantified.
The technique used by the gymnast or tumbler during the takeoff phase is also clearly
important for a successful performance with gymnasts spending years learning the techniques
required to perform a given tumbling movement. However, little is known about the
relationship between muscle activation timings and subsequent performance.
The purpose of this paper is to investigate how approach characteristics and takeoff
technique affect performance with a view to maximising somersault rotation in tumbling.

METHOD

A computer simulation model of the takeoff phase in tumbling was developed and
customised to an elite gymnast through the determination of subject specific inertia and
strength parameters. A simulation was produced which matched an actual performance of a
double layout somersault by the elite gymnast. The simulation model was then used to
maximise somersault rotation by varying the technique used during the takeoff phase for three
sets of initial conditions at touchdown with the tumbling track.
Ninety-five anthropometric measurements of the elite gymnast were taken and
segmental inertia parameters were calculated using the mathematical model of Yeadon
(1990b). One double layout somersault by the gymnast was recorded using a Locam 16mm
cine camera operating at 200 Hz and two 50 Hz Hi8 video cameras. The Locam and one
video camera were oriented perpendicular to the tumbling track and the other video camera
was positioned behind the landing area. Fifteen body landmarks (wrist, elbow, shoulder, hip,
knee, ankle and toe on both sides of the body plus the centre of the head) were digitised
throughout the movement from both camera views. Quintic splines were fitted to the
digitised data (Wood and Jennings, 1979). The closeness of fit at each point was based on the
difference between the data and a pseudo data set that was generated by averaging the
digitised data from the two adjacent times. DLT reconstructions (Abdel-Aziz and Karara,
1971) were then carried out to synchronise the digitised data (Yeadon and King, 1999) and
obtain 3D co-ordinate time histories of each body landmark at 0.005 s time intervals for the
takeoff phase and 0.020 s time intervals for the flight phases. The 3D data were then used to
calculate orientation and configuration angles (Yeadon, 1990a) which were fitted with quintic
splines (Wood and Jennings, 1979) in order to obtain angle and angular velocity estimates.
Error estimates were again obtained using pseudo data sets. The mass centre location was
calculated from the 3D data and the segmental inertia parameters of the gymnast. The mass
centre locations at the start and end of the flight phases before and after contact with the
tumbling track were used to determine the horizontal and vertical mass centre velocities at
touchdown and takeoff using equations of constant acceleration. The whole body angular
momentum about the mass centre at touchdown and takeoff from the tumbling track were
calculated as the mean angular momenta values during each flight phase (Yeadon, 1990c).

2
 Maximal isovelocity extension torque data were collected for the gymnast using a two
repetition concentric-eccentric protocol at preset crank angular velocities ranging from 20s-1
to 250s-1 for the ankle, knee, hip and shoulder using an KinCom 125E dynamometer. The
joint torque data obtained were fitted using an 18 parameter exponential function of angular
velocity and angle (King and Yeadon, 2002). The 18 torque parameters for each joint were
calculated by minimising the sum of squares of differences between the measured torque
values and the exponential function using Simulated Annealing (Goffe et al., 1994). The
gymnast gave informed consent for these procedures in accordance with the protocol
approved by the Loughborough University Ethical Advisory Committee.
A planar five-segment model consisting of a foot, shank, thigh, trunk + head, and arm +
hand segments was developed for simulating the takeoff phase of tumbling (Figure 1). The
elastic properties of the tumbling track were represented by massless damped linear springs
which applied horizontal and vertical forces at the toe and vertical forces at the ankle when
the toe and/or heel were in contact with the tumbling track. The model had four torque
generators Ta, Tk, Th and Ts (Figure 1) which opened (increased) the ankle, knee, hip and
shoulder joint angles aa, ka, ha and sa (extension at the ankle, knee and hip; flexion at the
shoulder). At a given moment in time the torque exerted at a joint was the product of the
maximum attainable torque at given joint angle / angular velocity and the torque activation
level at that time (between zero and one). Each torque generator was allowed to have an
initial torque value corresponding to a maximum of 50% of full activation and to remain at
this level for a period of time before ramping up to the final level (less than or equal to full
activation). The ramping function increased from zero to the final level over a time period
greater than or equal to 50 ms (King and Yeadon, 2003). A rotational elastic component with
a stiffness value of 465 Nm.rad-1 was included in series with the torque generator at the ankle
joint. This stiffness value was based upon an elastic element of length 0.314 m in the muscle-
tendon complex of the tricep-surae muscle group with a moment arm of 0.046 m (Jacobs et
al., 1996) and a maximum stretch of the elastic element of 4% at maximum torque (Bobbert
and van Ingen Schenau, 1990).

Figure 1. The five-segment simulation model of tumbling takeoff. Four torque generators (Ta, Tk, Th, Ts) open
the ankle, knee, hip and shoulder joint angles aa, ka, ha and sa and two springs allow for horizontal and
vertical movement of the tumbling track.

The FORTRAN code implementing the model was generated using the Autolev
software package which is based on Kane’s method of formulating the equations of motion
(Kane and Levinson, 1985). Subject specific model parameters comprised the previously
3
determined segmental inertias and joint torque parameters. Input to the simulation model
comprised the motion of the system just prior to touchdown of the model with the tumbling
track (mass centre velocity, orientation of each segment, angular velocity of each segment)
and for each torque generator: the initial activation, the onset time, the ramp time and the final
(greatest) activation level. The output from the model comprised whole body angular
momentum about the mass centre, mass centre velocity, orientation and angular velocity of
each segment at the time of takeoff from the tumbling track. The simulated performance
during flight was then determined using a three-dimensional 11-segment model of aerial
movement (Yeadon et al., 1990) which used configuration angles as input.
A matching procedure was used to obtain a simulation that was in close agreement with
the recorded double layout somersault performance. The Simulated Annealing algorithm
(Goffe, et al., 1994) was used to obtain the best match by minimising a cost function which
was based on the difference between a simulation and the actual performance in terms of
strategy used (vals) and takeoff (valt). The strategy component consisted of the four joint
angles at takeoff, the trunk segment angle at takeoff and the minimum ankle and knee angles
during the takeoff phase. For the calculation of vals each joint (ankle, knee, hip and shoulder)
was given a weighting of 1/8 and the trunk angle was given a weighting of 1/2 (equal to the
total weighting of the joint angles) since the trunk angle represented the whole body
orientation whereas the joint angles defined the configuration. vals therefore measured the
difference in the strategy used between a simulation and the actual performance in degrees.
The takeoff component comprised the horizontal and vertical velocity of the mass centre and
the whole body angular momentum at takeoff. The weightings for each variable in valt were
set in proportion to the inverse of the value of each variable from the actual performance. The
effect of using these weightings was that valt represented the average percentage difference
between a simulation and an actual performance in terms of the velocity and angular
momentum at takeoff. The cost function for a simulation was then calculated by averaging
valt and vals since 10% for valt was considered to be comparable with 10 for vals.
The initial conditions for the matching simulation were estimated from the video
analysis of the actual performance and corresponded to the time of touchdown with the
tumbling track. The mass centre velocity and the segment angles were fixed at the values
estimated from the video analysis as these were considered to be sufficiently accurate. The
five initial segment angular velocities, however, were allowed to vary by 50/s in the
matching optimisation as these estimates were not considered to be very accurate. In addition
20 other parameters were varied in the matching optimisation. Sixteen of these specified
technique by defining the activation time histories of the four torque generators (initial
activation, the time that the activation changes from the initial level, the ramp time and the
final (greatest) activation level) and four parameters governed the characteristics of the elastic
tumbling track. The Simulated Annealing algorithm (Goffe et al., 1994) was used to vary the
25 parameters until the cost function was minimised and the best match was found. The
optimisation routine was run with different initial conditions for the parameters to guard
against the routine becoming stuck in a local minimum. A typical optimisation evaluated up
to 20,000 simulations and took 24 hours to run.
Values for the horizontal approach velocity and whole body angular velocity at the start
of the takeoff phase were calculated for an elite male tumbler who performed one double
straight somersault and one triple twisting double straight somersault. The maximum values
obtained were then increased by 10% to give realistic limiting values for the horizontal
velocity and angular momentum at touchdown.
Three optimisations were carried out to maximise rotation potential (flight time 
angular momentum at takeoff) for various initial conditions. The flight time was constrained
4
to be greater than 1.05 s (that of the actual double layout somersault performance) to ensure
that the optimum simulations were not unrealistically fast and low. For each optimisation 21
parameters were varied, 16 of which defined the activation time histories for the four torque
generators, while the remaining five parameters defined the initial body configuration and
orientation. The difference between the three optimisations was in the initial linear and
angular momenta (at touchdown) input to the simulation model. For the first optimisation
(Optimisation 1) the linear and angular momenta at touchdown were fixed at the values
obtained for the actual double layout performance. For the second optimisation (Optimisation
2) the initial horizontal velocity at touchdown was allowed to vary (up to the limiting value)
and in the third optimisation (Optimisation 3) the initial angular momentum at touchdown
was allowed to vary as well (up to the limiting value). In addition to the three optimisations,
two single simulations were performed to establish the effect of changing the initial
conditions without re-optimising the technique used. The first single simulation used the
technique from Optimisation 1 with the optimised horizontal velocity at touchdown from
Optimisation 2. The second single simulation used the technique from Optimisation 2 with
the optimised angular momentum at touchdown from Optimisation 3. The simulation model
of Yeadon et al. (1990) was used with the results of each simulation of the takeoff phase to
determine how much somersault rotation could be achieved during the flight phase. Two
configuration strategies were used during the flight phase, one corresponding to the layout
configuration used by the gymnast in the actual double layout somersault and the other
corresponding to a straight configuration with the arms by the sides and the body extended.
The lower moment of inertia for the layout configuration permits more rotation than the
straight configuration.

RESULTS AND DISCUSSION

Figure 2. Comparison of actual performance and matching simulation of a double layout somersault (the gymnast
approaches from the left with a backward handspring).

By optimising the initial segment angular velocities (Table 1), the activation timings
(Table 2) and the stiffness / damping parameters for the tumbling track (horizontal stiffness:
131,361 Nm-1, vertical stiffness: 56,732 Nm-1, horizontal damping: 0 Nsm-1, vertical damping:
148 Nsm-1) close agreement was obtained between the actual double layout somersault
5
performance and the matching simulation (Figure 2). The average difference in linear and
angular momenta at takeoff from the tumbling track was less than 1%, and the average
difference in segment angles at takeoff was less than 1 (Table 3). It would therefore appear
that using a simple activation profile defined by four parameters is able to approximate the
activation profile at each joint. In the future it would be useful to compare the activations
timings used to EMG data collected during tumbling movements. More complex activation
profiles could have been used in the study and this may have improved the agreement in the
joint angle changes (Figure 3). This, however, would have required many more parameters to
be varied. The overall agreement between actual performance and the matching simulation
was considered to be sufficiently close to allow the simulation model to be used to investigate
how changing the approach characteristics and technique affects the production of rotation
potential during the takeoff phase.

Table 1. Initial conditions at touchdown for the matching double layout

simulation and the three optimisations

variable matching Opt 1 Opt 2 Opt 3

ug [ms-1] 4.83 4.83 7.00 7.00

vg [ms-1] -0.27 -0.27 -0.27 -0.27

aa 100 111 113 96

ka 144 146 145 141

ha 116 114 130 131

sa 148 135 125 140

tra 17 13 23 20

cma 57 54 51 47

a -821s-1 -821s-1 -821s-1 -821s-1

k -278s-1 -278s-1 -278s-1 -278s-1

h 715s-1 715s-1 715s-1 715s-1

s -157s-1 -157s-1 -157s-1 -157s-1

tr 931s-1 955s-1 924s-1 1005s-1

Note: ai, ki, hi, si and tri = the ankle, knee hip, shoulder and trunk angles (i = a) and angular
velocities (i = ) at touchdown; ug and vg = the horizontal and vertical velocity of the mass
centre at touchdown. cma = the angle of the mass centre to toe line relative to the horizontal
at touchdown (body orientation at touchdown). The trunk angle tra is the angle the trunk
makes with the horizontal and the joint angles are shown in Figure 1.

Using torque generators is a potential limitation of the study as the effect of biarticular
muscles are not completely accounted for when determining the velocity of shortening during
a simulation. However using torque generators does allow subject specific parameters to be
determined which includes the torque produced by biarticular muscles. Since the model has
6
previously been evaluated (Yeadon and King, 2002) and the matching in the current study is
good, the effect of neglecting to model biarticular muscles specifically is assumed to be small.
The inclusion of a series elastic element at the ankle joint has previously been shown to
improve the agreement between actual performance and simulation by less than 2% (Yeadon
and King, 2002). Thus any errors arising from the use of data from the literature for the
amount of stretch in the series elastic element and the lengths of the contractile and elastic
elements will have small effect and will not affect the findings of the study.
An elite tumbler was found to have a horizontal velocity of approximately 6.4 ms-1 and
a whole body angular velocity of approximately 720s-1 at touchdown when performing a
double straight somersault and a triple twisting double straight somersault. Limiting values
for the approach velocity and angular velocity were estimated at 7.0 ms-1 and 800s-1. These
were considered to be achievable limiting values in that they were 10% higher than the
measured velocities.

Figure 3. Comparison of key kinematic variables during the takeoff phase; solid line = actual performance,
dashed line = matching simulation data.

Optimisation 1 demonstrated that it was possible to adopt suitable activation timings in

order to change the linear and angular momenta at takeoff and produce a double straight
somersault (Figure 4). Optimisation 1 had lower activation levels at the knee and greater
activation levels at the hip and shoulder compared to the matching simulation (Table 2)
resulting in greater knee flexion at takeoff (Table 3). In addition the orientation of the body at
touchdown was 3 lower than the matching simulation (Table 1). The optimum simulation
had sufficient height and angular momentum to permit a straight body configuration during
flight and to land on the tumbling track. This was a significant improvement (19% more
rotation potential) on the actual double layout somersault performance which landed on a mat
in a foam filled landing pit. The total energy at takeoff for the optimum solution was 1%
7
more than in the matching simulation which was 4% more than the energy at touchdown. It
would therefore appear that optimising the activation profiles is able to change the
distribution of linear and angular momenta during takeoff but does not have a large effect on
the total energy of the system. As a consequence, maximising somersault rotation appears to
require high initial kinetic energy at touchdown along with suitable activation timings during
the takeoff phase.

Table 2. Activation parameters for the four torque generators in the matching
double layout simulation and the three optimisations

parameter matching Opt 1 Opt 2 Opt 3

ia 5% 5% 5% 5%

ik 11% 14% 11% 38%

ih 34% 48% 48% 50%

is 18% 50% 49% 50%

ta 0.118 s 0.077 s 0.072 s 0.066 s

tk 0.074 s 0.157 s 0.283 s 0.135 s

th 0.034 s 0.027 s 0.027 s 0.025 s

ts 0.281 s 0.415 s 0.025 s 0.025 s

ra 0.103 s 0.050 s 0.050 s 0.050 s

rk 0.063 s 0.070 s 0.180 s 0.120 s

rh 0.057 s 0.051 s 0.051 s 0.050 s

rs 0.261 s 0.372 s 0.050 s 0.050 s

maxa 100% 100% 100% 100%

maxk 91% 98% 91% 100%

maxh 100% 100% 100% 100%

maxs 100% 100% 100% 100%

Note: ij = the initial activation expressed as a percentage of full activation for ankle
(j=a), knee (j=b), hip (j=h), shoulder (j=s), tj = the time that the activation
reaches final (greatest) level, rj = the corresponding ramp time and maxj = final
activation level reached in a simulation as a percentage of full activation.

Optimisation 2 was used to determine the optimum horizontal approach velocity for
maximising somersault rotation. The optimum solution was found to use the maximum
permitted horizontal velocity of 7.0 ms-1 (Table 1) and with a consonant technique during the
takeoff phase (Table 2) produced sufficient rotation potential (31% increase above
Optimisation 1) to allow a triple layout somersault to be produced with the landing into a pit
9Figure 4). Without the changes in activation timings and body configuration and orientation
8
at touchdown the increase in approach velocity had a detrimental effect with a 42% reduction
in the rotation potential at takeoff of Optimisation 1. This demonstrated that an increase in
the horizontal velocity at touchdown is only beneficial if it is accompanied by the appropriate
activation timings. The differences in the activation profiles between Optimisation 1 and
Optimisation 2 were at the knee and shoulder where the knee activation profile was reduced
and the shoulder was activated maximally (Table 2) resulting in greater knee flexion at
takeoff. These differences allow more angular momentum to be produced during the takeoff
phase. The initial orientation of the body (at touchdown) was 3 lower than in Optimisation 1
(Table 1). The total energy at takeoff was within 1% of the energy at touchdown, with the
total energy approximately 40% greater at touchdown and takeoff when compared with the
double layout performance. Therefore a triple layout somersault requires a much faster
approach with more energy generated and appropriate activation timings during the takeoff
phase. As a consequence a triple layout somersault could not be performed in Artistic
Gymnastics in the floor exercise where the restricted run-up results in approach velocities of
around 4.5 ms-1 (Hwang, Seo and Liu, 1990). An additional optimisation was carried out with
the upper bound for the horizontal approach velocity removed and an optimum approach
velocity of 10.7 ms-1 was found beyond which the knees started to collapse during the takeoff
phase.
Table 3. Comparison of the double layout performance with the matching
simulation and the three optimisations

variable actual matching Opt 1 Opt 2 Opt 3

ug [ms-1] 2.48 2.51 1.91 3.59 2.97

vg [ms-1] 4.71 4.72 4.72 5.27 5.62

hg [kgm2.rads-1] 96 97 116 138 143

aamin 74o 73o 73o 73o 71o

aa 125o 126o 133o 119o 111o

kamin 142o 139o 131o 130o 119o

ka 168o 169o 142o 136o 123o

ha 202o 202o 180o 175o 177o

sa 153o 153o 168o 163o 168o

tra 99o 98o 92o 94o 92o

Note: For strategy component vals: aamin and kamin = the minimum ankle and knee angles; aa, ka, ha,
sa and tra = the ankle, knee hip, shoulder and trunk angles at takeoff. For takeoff
component valt: ug and vg = the horizontal and vertical velocity of the mass centre at
takeoff; hg = the angular momentum about a transverse axis through the mass centre at
takeoff.

Optimisation 3 found that having greater whole body angular momentum at touchdown
(18% increase) was beneficial for the maximisation of somersault rotation so long as a
suitable technique could be adopted. The optimum solution possessed 9% more rotation
potential than Optimisation 2 and this was sufficient to allow a triple straight somersault to be
performed with the landing on the tumbling track (Figure 4). Without the changes in
9
activation timings and body configuration and orientation at touchdown the increase in
angular momentum at touchdown had a detrimental effect with a 4% reduction in the rotation
potential at takeoff when compared with Optimisation 2. This demonstrated that an increase
in the angular momentum at touchdown is only beneficial if it is accompanied by the
appropriate technique. The initial orientation of the body (at touchdown) was 4 lower than in
Optimisation 2 (Table 1) and the major difference in the activation timings was at the knee
where the activation level was higher. This is probably due to the increased loading on the
knee (higher initial angular momentum) requiring a higher knee torque to ensure that the knee
extends before takeoff. Again the energy at touchdown and takeoff were similar and were
approximately 50% greater than the energy of the actual double layout somersault
performance.

Figure 4. Graphics sequences showing the optimum simulations for maximising rotation with (1) original
approach velocities (2) horizontal approach velocity of 7 ms-1 (3) angular approach velocity of 800s-1
and horizontal approach velocity of 7 ms-1.

A simple maximisation of somersault rotation was used as an optimisation criterion in

this study and the conditions for producing a triple somersault with a straight body were
found. It may be concluded that the limiting factor to maximising somersault rotation is the
ability to generate high linear and angular velocities during the approach phase coupled with
the ability to adopt consonant activation timings during the takeoff phase. In practice a
gymnast will adopt an optimisation strategy that produces a robust solution that is also close
to optimal in the above narrow sense of maximising rotation potential. In other words small
errors in technique will produce only a small reduction in performance. In future analyses
such considerations should be included in optimisation criteria since there is evidence that
robustness can explain technique in, for example, targeted movements (Harris and Wolpert,
1998).

10
Acknowledgments

The financial support of the Biotechnology and Biological Sciences Research Council
(BBSRC) and the cooperation of British Gymnastics are gratefully acknowledged.

References

Abdel-Aziz, Y.I., Karara, H.M., 1971. Direct linear transformation from comparator
coordinates into object-space coordinates. (ASP Symposium on Close-Range
Photogrammetry). Falls Church, VA: American Society of Photogrammetry, pp. 1-18.
Bobbert, M.F., van Ingen Schenau, G.J., 1990. Isokinetic plantar flexion: Experimental
results and model calculations. Journal of Biomechanics 23, 105-119.
Brüggemann, G.P., 1983. Kinematics and Kinetics of the backward somersault take-off from
the floor. In H. Matsui & K. Kobayashi (Eds.), Biomechanics VII-B. Champaign, IL:
Human Kinetics, pp. 793-800.
Brüggemann, G.P., 1987. Biomechanics in gymnastics. In B. van Gheluwe & J. Atha (Eds.),
Current research in sports biomechanics. Medicine and Sport Science. Basel: Karger,
pp. 142-176.
Goffe, W.L., Ferrier, G.D., Rogers, J., 1994. Global optimisation of statistical functions with
simulated annealing. Journal of Econometrics 60, 65-99.
Hwang, I., Seo, G., and Liu, Z.C., 1990. Takeoff mechanics of the double backward
somersault. International Journal of Sport Biomechanics, 6, 177-186.
Jacobs, R., Bobbert, M.F., van Ingen Schenau, G.J., 1996. Mechanical output from individual
muscles during explosive leg extensions: The role of biarticular muscles. Journal of
Biomechanics 29, 513-523.
Harris, C.M. and Wolpert, D.M. 1998. Signal-dependent noise determines motor planning.
Nature 394, 780-784.
Kane, T.R., Levinson, D.A., 1985. Dynamics: Theory and applications. New York:
McGraw-Hill Book Company.
King, M.A., Yeadon, M.R., 2002. Determining subject specific torque parameters for use in a
torque driven simulation model of dynamic jumping. Journal of Applied Biomechanics,
18, 207-217.
King, M.A., Yeadon, M.R., 2003. Coping with perturbations to a layout somersault in
tumbling. Journal of Biomechanics 36, 921-927.
Wood, G.A., Jennings, L.S., 1979. On the use of spline functions for data smoothing. Journal of
Biomechanics 12, 477-479.
Yeadon, M.R., 1990a. The simulation of aerial movement – I. The determination of
orientation angles from film data. Journal of Biomechanics, 23, 59-66.
Yeadon, M.R., 1990b. The simulation of aerial movement – II. A mathematical inertia
model of the human body. Journal of Biomechanics, 23, 67-74.
Yeadon, M.R., 1990c. The simulation of aerial movement – III. The determination of the
angular momentum of the human body. Journal of Biomechanics, 23, 74-83.
Yeadon, M.R., Atha, J., Hales, F.D., 1990. The simulation of aerial movement – IV. A
computer simulation model. Journal of Biomechanics, 23, 85-89.
Yeadon, M.R., King, M.A., 1999. A method for synchronising digitised video data. Journal
of Biomechanics, 32, 983-986.
Yeadon, M.R., King, M.A., 2002. Evaluation of a torque driven simulation model of
tumbling. Journal of Applied Biomechanics, 18, 195-206.
11