This repository contains MATLAB code and data files for analyzing the biomechanics of turning in guinea fowl (Numida meleagris) across high and low friction terrain conditions. The analysis investigates how birds modulate ground reaction forces, limb posture, and body center of mass dynamics during straight runs and turns on high-friction (control) and low-friction (slippery) surfaces. The files contain ground reaction forces, center of pressure, and body center of mass (CoM) trajectories from guinea fowl as they navigated straight and 90-degree turning runways,  on high and low friction surfaces.  The data and code are associated with the paper: “Faster fowl fall frequently: speed and force regulation during turning maneuvers by guinea fowl on high and low friction terrains”, and all the specific details of the methods are explained in further detail in that paper. 

[These methods are from the associated manuscript, therefore references to tables, figures and supplemental information are for the associated manuscript.  However, the dataset and code here fully reproduces the statistical analysis described.  See the README file]

Animals and training: Seven helmeted guinea fowl (Numida meleagris, 4 males, 3 females) were used in this study, which were hatched and reared at the Royal Veterinary College (RVC). Sample size is based on effect size estimates from previous studies on ostriches turning and guinea fowl on slippery terrain (Jindrich et al. 2007; Clark and Higham 2011). Eight birds were initially trained, but one animal was excluded from the analysis because it refused to cross slippery terrain and therefore had an incomplete dataset. At the time of testing, the guinea fowl were 4 years old with body mass 1.89 ± 0.21 kg and standing hip height 17.53 ± 2.38cm (mean ± s.d). Primary feathers were kept clipped to the length of the secondary feathers to prevent flying. Birds were housed as a group and tagged with colored cable ties on the distal tarsometatarsus for identification. Training and experimental data collection were conducted at the RVC and ethical approval was granted by the RVC Clinical Research Ethical Review Board (CRERB: Protocol # URN: 2017 1759-3 to M.A.D).

Terrain conditions and runway construction: We recorded trials on four runway conditions: straight and 90° turns, each with high- and low-friction surfaces. These four conditions were designated as control straight (CS), slippery straight (SS), control turns (CT) and slippery turns (ST). Low-friction terrain was created using smooth polypropylene sheeting, which was established by Clark and Higham to successfully elicit a substantial increase in slipping frequency in guinea fowl (Clark and Higham 2011). The control, higher friction substrate was gritted anti-skid paint specifically designed to increase traction. The runway arena was constructed from plywood, Rexroth metal rods and fittings, and transparent Perspex to allow camera views for high-speed video recordings. The runway was 1m high, designed to be higher than a guinea fowl’s typical maximal jumping height when not aided by flight (~0.8 m, Henry et al. 2005). The runway was constructed around an array of six 0.9m x 0.6m force plates (Kistler type 9287BA, Amherst, NY, United States), oriented with the long side spanning the width of the runway. The straight runway was 1m wide x 9m long, with a 3.6m section in the middle containing the six force plates. In the turning runway, a 90° turn occurred at the last 0.9m of the 3.6m section formed by the sequence of six force plates. The straight section after exiting the corner continued for 2.7m, forming an ‘L-shape’. In low friction, slippery terrain trials, the entire runway surface was covered with polypropylene sheeting cut to fit the space, with sections over the force plates cut to the exact dimensions of the force platform top plates. At each end of the runway, dark black boxes were adjoined to the runway arena, to allow rest, food and water between trials. The width of the turning runway was wide enough to allow the bird to select a range of paths and turn sharpness as part of its locomotor strategy. This design was intentional, to allow comparison of turning strategy between conditions, to explore how birds prioritize physical demands and performance trade-offs that emerge between speed, turning and terrain friction.

Two AOS high speed video cameras were positioned facing the runway (S-PRI, Technologies AG, Dättwil, Switzerland), aligned to capture sagittal plane kinematics as the bird ran from one end of the runway to the other. In the turning runway, the cameras were positioned at 90° angles, with overlapping views in the corner where turns were executed. Three infrared LED lamps illuminated the runway arena to ensure good video image quality.

Bird Training: Pilot testing trials were performed to acclimatize each bird to the laboratory environment and runway locomotion; however, the birds were not exposed to the turning or slippery terrain conditions before data collection. During training, birds were first allowed to rest in a dark box at the end of the runway with food and water. After a few minutes of quiet rest, a vertical sliding door was opened at the front of the box, and the bird encouraged (using waving, clapping and visual motion of looming objects) to run to the opposite end of the runway to another dark box. Once the bird reached the box at the opposite end of the runway, a sliding door was closed, and mealworms provided a positive food reward. The bird was then transferred to the starting box and allowed to rest for several minutes. Through this process, the birds learned to locomote continuously across the runway section to reach the resting boxes at either end.

Experimental protocol: The birds were exposed to repeated trials of four runway conditions; control straight (CS), slippery straight (SS), control turns (CT) and slippery turns (ST). Trials were collected over a 2-month period in the same sequence for each bird: CT, ST, SS, CS. Consistent sequencing among birds was selected in part for practical reasons because changing the runway configuration was very labor intensive. Considering that the control straight (CS) condition was collected last, there could be some carry over effects from exposure to the manipulated terrain conditions. However, this is expected to lead to conservative estimates for the differences between terrains.

Birds were brought in as pairs each testing day for companionship. While one bird completed trials, the other was placed in a carrier next to the box at the end of the runway to provide additional motivation to reach the resting box. Birds were placed in the starting black box and a small number of practice trials were performed for habituation, until the birds were able to move continuously across the runway from one box to the other. To begin each trial, the cover on the starting box was lifted and the gate was opened. Force and video were synchronously triggered to record for 10s as the bird emerged from the starting box. When the bird reached the end box it was shut, covered, and the bird was transferred back to the starting black box and left to recover for several minutes.

Inclusion & Exclusion Criteria: A successful trial was achieved when the bird was able to move continuously across the runway and remain within the runway without contacting the walls, and without excessive wing-flapping or jumping. Each bird varied in both its willingness to participate in continued trials and the consistency of behavior within trials. Trials were repeated until we collected at least 10 good continuous locomotion trials, but we stopped data collection if the bird became visibly tired, distressed or uncooperative. More than 10 trials were collected when possible, and in these cases, we recorded all trials but sampled a subset of 10 trials that were the most continuous and fast trials for that bird, without wall contact events. There were only two cases where we collected the minimum of 10 trials, both cases were with different individuals on two different runway conditions. For these cases, we could not institute any selective criteria, and all 10 trials were used.

Digitizing: High-speed video was recorded at 160fps. For each trial, we tracked the sagittal plane kinematics of the approximate body center of mass (CoM) along the initial running direction, using DLTdv5 digitizing program (Hedrick 2008), which runs in Matlab (Mathworks, Natick, Massachusetts). An initial estimate of the body CoM position was obtained by manually tracking a point on the body located approximately 50% along the dorso-ventral axis, and 40% along the cranial-caudal axis. This relative position of the guinea fowl body CoM was estimated using the suspension technique as described by Fedak and colleagues (Fedak et al. 1982). The digitized trials were calibrated and corrected for lens distortion and parallax by digitizing the corners of the six force plates that spanned the runway. The digitized body CoM position data were used to estimate the initial condition values (position and velocity) required to double-integrate acceleration from the force plate data to calculate CoM trajectory over time. These initial condition estimates were then refined using the path-match optimization algorithm that minimizes the drift over time of the kinematic versus force-plate derived estimate of CoM position over the course of the whole trial (Daley et al. 2006; Blum et al. 2014).

Videos were also manually tracked in ELAN software (version 5.2, 2018: https://archive.mpi.nl/tla/elan; Sloetjes and Wittenberg 2008) to identify slip, fall and collision events. A slip was defined as an event when the foot moved horizontally along the substrate while flat against the ground during midstance. A fall occurred when another body part, other than the foot, contacted the ground. A collision occurred when the body contacted a wall.

Data processing to calculate kinetics from ground reaction force data: Ground reaction force (GRF) and center of pressure (CoP) data were collected at a sampling frequency of 800Hz. All data processing was completed using custom scripts in Matlab (Mathworks, Natick, Massachusetts). Ground reaction force data were corrected for offset and drift by fitting a line to the baseline and subtracting this fitted line from the data. We then applied a 4^th-order low pass recursive Butterworth filter with a cutoff frequency of 50Hz. To reduce the noise inherent in CoP data at low forces, the CoP trajectories were smoothed using a linear smoothing spline to remove unrealistic transients at the start and end of stance.

Trials were trimmed to correspond to the time-period that the bird was on the force plate array. We then calculated the body center of mass (CoM) velocity and position by double integrating the acceleration data, where acceleration was calculated by dividing ground reaction force by body mass (Cavagna 1975; Daley et al. 2006).  Initial conditions for integration were found using the path match optimization technique initially described by Daley and colleagues (Daley et al. 2006). The path match optimization minimizes the sum of the squared error (SSE) between the kinematically estimated paths and the force-plate derived CoM paths, to avoid drift due to errors in the initial position and velocity estimates. The optimization function ‘fminsearch’ in Matlab was used to iteratively adjust the initial condition estimate to minimize the drift between the two trajectory estimates. This method allows within-stride deviations between the kinematic and force-plate derived trajectories. Deviations between these trajectories are expected because kinematic estimates are based on fixed anatomical landmarks, not the true CoM. The optimization minimizes the drift between the two trajectories over time, which results from error in initial condition estimates. We used the digitized kinematics for the fore-aft and vertical CoM trajectory estimates but used the smoothed CoP trajectory to estimate the medio-lateral initial conditions, because the video was only digitized for the sagittal plane. However, initial conditions for the medio-lateral direction are expected to be close to zero because the birds initially run in a straight line. Our final CoM trajectory calculations are based on the integrated 3D ground reaction forces, enabling us to measure changes in velocity heading in three dimensions over the course of the trial. We conducted a check on the maximum deviation between the kinematic and kinetic CoM trajectories and flagged any trials with a maximum deviation greater than 7cm. These trials typically involved instances where the bird's body touched a surface other than the force plate, and the trials were trimmed to analyze the section of the trial where no such instances occurred.

Step detection and variables measured: Before segmenting data into steps, the position, velocity, force and acceleration trajectories were converted from a lab reference frame into an anatomical reference frame based on the velocity heading, such that the fore-aft direction always corresponds to the current direction of travel at the start of each stance phase. Data were then segmented into step cycles using minimum detection on the fore-aft acceleration trajectory, which we found to be the best proxy for touchdown across both walking and running speeds. This touchdown proxy was cross-checked against step counts taken from the video data. We initially used automated peak detection using the ‘findpeaks’ function in Matlab, and then visually checked the peak detection through a user-interface that allowed iterative adjustment of the minimum peak prominence and minimum number of points between peaks until the algorithm detected a single fore-aft minimum per step.

Once the trials were segmented into step cycles, we measured the following variables at for the start and end of each step period: the position and velocity of the body CoM, the position of the foot center of pressure (CoP), and the effective length and angle of the virtual leg (calculated from the line connecting the body CoM to the foot CoP). From these measures, we were able to calculate initial conditions at the start of each step and the net change within each step, including the change in velocity magnitude and heading. We also measured step length, step duration, the force impulse in each direction, the minimum, maximum and mean force in each direction, the peak and mean magnitude of the resultant force vector (F_R).

Statistical analysis: All variables were normalized to dimensionless quantities before statistical analysis, based on standing hip height, body mass and gravity, according to the conventions of McMahon and Chang (McMahon and Cheng 1990; Daley and Birn-Jeffery, 2018). This accounts for differences in variables associated with variation in body size, assuming geometric and dynamic similarity between individuals. Visual inspection of the CoM trajectories revealed that birds typically initiated the turn after they reached about 1.5 meters along the runway. We therefore focused the statistical analysis on steps occurring after this position. We used a linear mixed-effects ANOVA with terrain condition as a main fixed factor, and individual subject as a random factor. We compared multiple models of varying complexity, and evaluated the models based on AIC and total adjusted R-squared. The null hypothesis model included only individual as a random effect, Model 1 (null hypothesis): Y ~ 1+ (1|subject). The first comparison models included condition as a fixed categorical factor, Model 2: Y~ 1+ condition + (1|subject). The sharpest turning steps tended to occur near the end of the runway, so we tested a model that included a linear covariate of runway position (pos_y, the birds progression along the runway, measured at the start of each step) as an interaction term with terrain condition: Model 3: (Y~ 1+ condition + conditionpos_y + (1|subject)). Finally, we also tested for learning effects with repeated trials by including trial number as a term nested within condition by individual: Model 4: Y~ 1+ condition + conditionpos_y + (condition:trialNum|subject). Models were compared based on AIC and adjusted R-squared, which supported selection of Model 4 based on AIC and adjusted R-squared (Table S1). We calculated F-statistics for the effects of condition, runway position and its interaction and coefficient estimate with 95% confidence intervals (CI). We used False Discovery Rate (FDR) correction to calculate an adjusted p-value threshold to maintain a 5% false positive rate across all statistical tests (Benjamini and Hochberg 1995). The F-statistics and linear mixed effects model outputs are provided in Table S2 and Table S3. The mean and 95% CIs for variables in each terrain condition were calculated based on marginal means after accounting for individual random effects, reported in Table 1A. We calculated pairwise marginal mean differences (and 95% CIs) for CT, SS and ST compared to the CS reference using the Dunnett procedure (Table 1B). The random effects term of (condition:trialNum|subject) resulted in a coefficient for the change in the variable of interest as a function of trial number within each treatment terrains, where CS is the reference terrain. We consider this coefficient to represent a learning effect of how locomotion changed with practice with repeated exposure to terrain conditions. These coefficients are reported for each individual in Table S4 in the associated manuscript.

Blum, Y., Vejdani, H. R., Birn-Jeffery, A. V., Hubicki, C. M., Hurst, J. W. and Daley, M. A. (2014). Swing-leg trajectory of running guinea fowl suggests task-level priority of force regulation rather than disturbance rejection. PLoS ONE 9, e100399. https://doi.org/10.1371/journal.pone.0100399

Cavagna, G. A. (1975). Force platforms as ergometers. J. Appl. Physiol. 39, 174-179. https://doi.org/10.1152/jappl.1975.39.1.174

Clark, A. J. and Higham, T. E. (2011). Slipping, sliding and stability: locomotor strategies for overcoming low-friction surfaces. J. Exp. Biol. 214, 1369-1378. https://doi.org/10.1242/jeb.051136

Daley, M. A., Usherwood, J. R., Felix, G. and Biewener, A. A. (2006). Running over rough terrain: guinea fowl maintain dynamic stability despite a large unexpected change in substrate height. J. Exp. Biol. 209, 171-187. https://doi.org/10.1242/jeb.01986

Fedak, M. A., Heglund, N. C. and Taylor, C. R. (1982). Energetics and mechanics of terrestrial locomotion II. Kinetic energy changes of the limbs and body as a function of speed and body size in birds and mammals. J. Exp. Biol. 97, 23-40. https://doi.org/10.1242/jeb.97.1.23

Hedrick, T. L. (2008). Software techniques for two- and three-dimensional kinematic measurements of biological and biomimetic systems. Bioinspir. Biomim. 3, 034001. https://doi.org/10.1088/1748-3182/3/3/034001

Henry, H. T., Ellerby, D. J. and Marsh, R. L. (2005). Performance of guinea fowl Numida meleagris during jumping requires storage and release of elastic energy. J. Exp. Biol. 208, 3293-3302. https://doi.org/10.1242/jeb.01764

Jindrich, D. L., Smith, N. C., Jespers, K. and Wilson, A. M. (2007). Mechanics of cutting maneuvers by ostriches (Struthio camelus). J. Exp. Biol. 210, 1378-1390. https://doi.org/10.1242/jeb.001545

McMahon, T. A. and Cheng, G. C. (1990). The mechanics of running: how does stiffness couple with speed? J. Biomech. 23 Suppl. 1, 65-78. https://doi.org/10.1016/0021-9290(90)90042-2

Sloetjes, H. and Wittenburg, P. (2008). Annotation by category-ELAN and ISO DCR. In 6th international Conference on Language Resources and Evaluation (LREC 2008). https://archive.mpi.nl/tla/elan\