Bearded capuchin monkeys use joint synergies to stabilize the hammer trajectory while cracking nuts in bipedal stance

Abstract

The transition from occasional to obligate bipedalism is a milestone in human evolution. However, because the fossil record is fragmentary and reconstructing behaviour from fossils is difficult, changes in the motor control strategies that accompanied this transition remain unknown. Quadrupedal primates that adopt a bipedal stance while using percussive tools provide a unique reference point to clarify one aspect of this transition, which is maintaining bipedal stance while handling massive objects. We found that while cracking nuts using massive stone hammers, wild bearded capuchin monkeys (Sapajus libidinosus) produce hammer trajectories with highly repeatable spatial profiles. Using an uncontrolled manifold analysis, we show that the monkeys used strong joint synergies to stabilize the hammer trajectory while lifting and lowering heavy hammers. The monkeys stringently controlled the motion of the foot. They controlled the motion of the lower arm and hand rather loosely, showing a greater variability across strikes. Overall, our findings indicate that while standing bipedally to lift and lower massive hammers, an arboreal quadrupedal primate must control motion in the joints of the lower body more stringently than motion in the joints of the upper body. Similar changes in the structure of motor variability required to accomplish this goal could have accompanied the evolutionary transition from occasional to obligate bipedalism in ancestral hominins.

1. Introduction

Comparisons of limb morphology and function among humans, extinct hominins and extant non-human primates demonstrate that during the evolutionary transition from occasional to obligate bipedalism, the feet and legs underwent more significant changes than the hands and arms [1,2]. However, because of the fragmentary nature of the fossil record and difficulties in reconstructing behaviour from fossils, changes in the motor control strategies that accompanied this transition remain unknown. Although supporting evidence is sparse, an accepted proposition is that altered motor control of the feet and legs that resulted in a progressive reduction in the displacement of the body's centre of mass (COM) accompanied the evolution of obligate bipedalism in ancestral hominins [3–6].

Capuchin monkeys (Cebus and Sapajus spp.), arboreal platyrrhines, spend a significant proportion of time foraging on the ground [7,8]. Unusually among primates, they sporadically walk bipedally on the ground [8]. Some wild populations of bearded capuchin monkeys (Sapajus libidinosus) crack open nuts and other encased food items using naturally available stone hammers by placing the nuts on stone or log anvils on or near the ground (figure 1) [9,10]. The monkeys in our study population at Fazenda Boa Vista (FBV) often use massive hammers (greater than 1 kg; 50% of an adult female's body mass) for processing more resistant nuts [11,12]. We estimate that other non-human primates that use stone hammers typically use proportionally lighter hammers (chimpanzees: less than 20% of body mass [13]; long-tailed macaques: less than 10% of body mass [14]). Capuchin monkeys at FBV carry hammers bipedally to anvils [15] and stand bipedally while using hammers [16]. They presumably solve significant biomechanical and postural challenges in lifting and lowering massive hammers while cracking nuts in bipedal stance.

Figure 1.

A wild bearded capuchin monkey is striking an intact piaçava nut with a quartzite stone hammer. (Photo by Noemi Spagnoletti.)

The ancestors of capuchin monkeys diverged from catarrhines, and thus from hominids, long before hominins adopted obligate bipedal locomotion. Studies of locomotion in apes that occasionally locomote bipedally have aided our understanding of the origins of bipedal locomotion in hominins [17–19], but it is also useful to consider other species and other bipedal activities in extant non-human primates for insights into the evolution of obligate bipedalism in ancestral hominins. For example, capuchin monkeys stand bipedally to crack nuts. These monkeys afford an opportunity to study motor strategies that support stable bipedal postures during strenuous action. Thus, although capuchin monkeys do not represent a progressive step in the evolution of bipedalism in ancestral hominins, they offer an independently evolved comparative reference point that is relevant to hominins' postural control in bipedal stance, as well as to postural control in bipedal locomotion with and without carrying a load in the arms [15,20–22].

In the present study, we identified the motor strategies bearded capuchin monkeys use to stabilize the hammer trajectory while cracking nuts in bipedal stance with massive hammers. We hypothesized that if the requirement of maintaining a bipedal stance poses significant biomechanical and postural challenges, then to stabilize the hammer trajectory, capuchin monkeys would more closely control motion in the joints of the lower body compared with the joints of the upper body. To test this hypothesis, we analysed the kinematics of striking movements performed by five wild adult monkeys. The monkeys struck intact palm nuts on a log anvil with hammers of different masses. We first examined repeatability in the spatial profiles of hammer trajectories and then used an uncontrolled manifold (UCM) analysis to determine whether and how the monkeys control motion differently in the joints of the lower and upper body.

The UCM analysis links trial-to-trial variability in the space of effector-level elemental variables or the degrees of freedom (DoFs) with variability in the task-relevant performance variables [23–26]. The concept of the UCM analysis is mostly used in the context of muscle synergies: multiple muscles work as functional units such that the central nervous system (CNS) jointly and proportionally activates all muscles in the synergy. When task demands change, the CNS control changes, resulting in changes in muscle synergies. By extending the notion of muscle synergies to ensembles of muscles that span multiple joints, we can understand the coordination of multiple joints. The UCM analysis proceeds by partitioning trial-to-trial variability in the space of effector-level elemental variables into two subspaces: controlled and uncontrolled. Variability in the controlled subspace influences the performance variable; variability in the uncontrolled subspace leaves the performance variable unchanged. A greater magnitude of variability in the uncontrolled subspace compared with the controlled subspace implies a synergy (figure 2a,b). The ratio of variability in the uncontrolled subspace to that in the controlled subspace reflects the strength of the synergy: stronger (figure 2c) or weaker (figure 2d). We examined (i) whether the monkeys structure variability in joint configurations in the two subspaces to minimize variability in the hammer trajectory, and (ii) how the monkeys control the DoFs of the lower and upper limbs while lifting and lowering massive hammers.

Figure 2.

Schematic illustration of the uncontrolled manifold (UCM) concept/analysis. (a) Not a synergy. Variability in the elemental variables in the controlled subspace, V_UCM, is smaller than variability in the uncontrolled subspace, V_ORT, that is, R_V = V_UCM/V_ORT < 1. (b) A synergy. V_UCM is greater than V_ORT, that is, R_V = V_UCM/V_ORT > 1. The magnitude of R_V reflects the strength of the synergy. (c) A stronger synergy. (d) A weaker synergy.

Capuchin monkeys use seven body joints (eight including the angle between the feet and the ground) to stabilize the hammer's horizontal and vertical positions [27], rendering the space of effector-level elemental variables a six-dimensional uncontrolled manifold. We anticipated that the monkeys—all proficient nut-crackers [12,28]—would use joint synergies to exploit this redundancy in movement space to stabilize the hammer trajectory. We previously found that spatio-temporal coordination between any two joints increases with hammer mass [27]. A higher degree of coordination implies fewer motor solutions, and consequently lesser redundancy in the movement space. We thus expected that the strength of synergy would decrease with hammer mass. Capuchin monkeys predominantly rely on the movement of their hindlimbs (hip and knee) and their torso (lumbar) to lift and lower a hammer, and to a limited extent on the movement of their forelimbs (shoulder) to lift a hammer [27]. We thus predicted that the monkeys would differently control motion in the joints of the lower and upper body.

2. Material and methods

(a). Subjects and study site

The subjects were five wild adult bearded capuchin monkeys (body mass: 2.1–4.3 kg) in their natural habitat at Fazenda Boa Vista (FBV) in Piauí, Brazil (9°39′ S, 45°25′ W; table 1). The monkeys at FBV crack open palm nuts during routine foraging [13]. In the present study, the monkeys cracked nuts of the piaçava palm, Orbignya spp., by placing each nut on a log anvil and striking it with a quartzite stone hammer. An intact piaçava nut is extremely resistant to fracture (mean ± s.d. peak force at failure = 11.50 ± 0.48 kN, n = 35), has a thick shell (i.e. the endocarp; thickness: 7.66 ± 0.30 mm, n = 35), and is a composite of several locules (mean ± s.d. number of locules: 3.00 ± 0.18, n = 35), each encapsulating a kernel (i.e. the endosperm) [29]. A piaçava nut also has an exocarp and an edible mesocarp that the monkeys themselves remove before cracking, or, more commonly at our site, grazing cattle remove them. Piaçava palm grows abundantly throughout FBV [29]. We collected the nuts locally. Their exocarps and mesocarps had already been removed by cattle or other animals. Log anvils and quartzite stones are naturally available at many locations at FBV, particularly near sandstone ridges [30]. We provided the monkeys with stones of three different masses: 1.01, 1.48 and 1.91 kg.

Table 1.

The number of striking movements analysed for each monkey.

			number of striking movements
monkey	sex	body mass (kg)	1.01 kg hammer	1.48 kg hammer	1.91 kg hammer
Mansinho	male	4.3	6	6	6
Teimoso	male	3.6	6	6	6
Presente	male	2.2	6	6	6
Dita	female	2.1	5	6	6
Chuchu	female	2.1	6	6	—

(b). Experimental procedure

We collected all data opportunistically. We placed a hammer on a log anvil and waited until a monkey voluntarily approached the anvil. We provided the approaching monkey with an intact piaçava nut by rolling the nut on the ground towards the anvil. We used a Canon XF100 HD camcorder (29.98 fps, 1920 × 1080 pi) mounted on a tripod at approximately 10 m from the anvil to record the monkey's actions. We captured each strike in the monkey's sagittal plane. We attached two physical markers 50 cm apart to the anvil to calibrate the plane of movement. We measured each monkey's body mass when it voluntarily stood on a digital scale mounted on a tree (details in [31]).

(c). Data extraction

The monkeys at FBV take several strikes to extract the kernel(s) of an intact piaçava nut and often modulate subsequent strikes according to the outcome of the previous strike (i.e. effective versus ineffective) [12]. Consecutive strikes can thus vary significantly in amplitude or kinetic energy, particularly after the locules of the nut are separated. Therefore, to keep the task demands constant across strikes, we coded from video only the first strike for each nut processed by each monkey. We coded six strikes per monkey per hammer (with two exceptions: one monkey provided five strikes with the lightest hammer, and one monkey provided no strikes with the heaviest hammer; table 1).

Two laboratory assistants, A.M. and J.Y.H., manually coded each strike using an open-source motion analysis software, Kinovea (https://www.kinovea.org/). A.M. and J.Y.H. placed a digital marker (‘+’) on each of nine anatomical locations on the monkey's body (figure 3a; table 2) and obtained the x-, y-coordinates of each location to the nearest pixel. One of the two physical markers attached to the anvil served as the origin of the plane of movement. The coders then forwarded the video by a frame, repositioned each digital marker and obtained the x-, y-coordinates of that marker. They iterated this process for the entire strike (i.e. mean ± s.d. = 22.6 ± 2.3 frames). The coding was highly consistent both within and across the two coders. Comparison of the x-, y-coordinates for three strikes coded twice by each coder over 15 days revealed Cronbach's alphas of 0.99 and 0.99, respectively. Comparison of the x-, y-coordinates for three strikes across the two coders revealed Cronbach's alpha of 1.00.

Figure 3.

UCM analysis. (a) The anatomical locations of the digital markers constituting the kinematic chain of striking movement. (b) The DoFs of the forward kinematic model linking the hammer's x-, y-positions to the monkey's joint configurations.

Table 2.

Anatomical locations of the digital markers constituting the kinematic chain of striking movement.

marker	anatomical location
finger—INF	distal phalanx of the index finger
wrist—WRI	wrist bar on the thumb side
elbow—ELB	lateral epicondyle approximating the elbow joint axis
shoulder—SHO	acromioclavicular joint
anterior superior iliac spine—ASI	anterior superior iliac spine
thigh—THI	lower lateral 1/3 surface of the thigh, just below the swing of the hand
knee—KNE	lateral epicondyle of the left knee
heel—HEE	calcaneus at the same height above the plantar surface of the foot as the toe marker
toe—TOE	second metatarsal head, on the midfoot side of the equinus break between forefoot and midfoot

(d). Data reduction

Each strike lasted mean ± s.d. = 0.75 ± 0.08 s. We divided each strike into two parts: the lifting phase up to the zenith and the lowering phase from the zenith to the end, using MatLab 2017b (MathWorks, Inc., Natick, MA, USA). Since the lifting phase was typically longer than the lowering phase, we re-sampled this time-normalized trajectory to 100 slices (50 slices for lifting and lowering each) through the cubic spline interpolation using the spline function in MatLab.

Changes in the x-, y-coordinates of a digital marker (the distal phalanx of the index finger, marker INF in figure 3a) constituted the hammer trajectory. For each monkey, we determined variability in the hammer trajectory along the horizontal and vertical axes (x- and y-axes, respectively) across all strikes. We shifted the hammer's x-, y-coordinates at the onset of each strike (i.e. t = 0) to the origin (i.e. x, y = 0). We then determined the standard deviation of the hammer's x-, y-positions in each of the 100 slices.

To perform the UCM analysis, we determined the eight joint elevation angles of each monkey constituting the kinematic chain of movement: foot (θ_foot), shank (θ_shank), thigh (θ_thigh), pelvis (θ_pelvis), trunk (θ_trunk), upper arm (θ_{upper arm}), lower arm (θ_{lower arm}) and hand (θ_hand) (figure 3b). Each elevation angle increased with counterclockwise rotation of the respective segment about the preceding segment in the kinematic chain.

(e). Uncontrolled manifold analysis

To begin the UCM analysis, we constructed a forward kinematic model—a set of equations—that allowed computing the hammer's x-, y-positions from the specified values of a monkey's joint configurations (figure 3b):

where x_m,hammer, y_m,hammer denote the hammer's x-, y-positions for the mth monkey (1 through 5), x_m,toe, y_m,toe denote the x-, y-positions of the mth monkey's toe, l_s denotes the average length of the mth monkey's sth segment (1 through 8), and θ_foot, θ_shank, θ_thigh, θ_pelvis, θ_trunk, θ_{upper arm}, θ_{lower arm} and θ_hand denote the DoFs constituting a strike.

We calculated the average joint vector, θ_m,t, for the mth monkey at the tth time slice (n = 24) averaged across all six strikes per hammer. Second, we calculated the difference between the average joint vector and the joint vector in the ith strike, that is, the deviation joint vector, \islant-20%\partialθ_m,t,i = θ_m,t − θ_m,t,i. Finally, based on the forward kinematic model, we calculated the Jacobian, J_m,t, for the mth monkey in the tth time slice:

This Jacobian is a linearized representation mapping infinitesimal changes across the DoFs onto the hammer's x-, y-positions. By calculating the null space of the Jacobian, J_m,tθ_j, we estimated the uncontrolled manifold (UCM) that contained V_UCM. All joint configurations specifying the same x-, y-positions of the hammer lay in this subspace. The subspace perpendicular to V_UCM represented the controlled subspace (ORT) and contained V_ORT. All joint configurations specifying different x-, y-positions of the hammer lay in this subspace. We then obtained a projection vector, V_UCM–m,t,i, for V_UCM for the mth monkey in the tth time slice during the ith movement by projecting the deviations vectors, \islant-20%\partialθ_m,t,i, onto the subspace of V_UCM. We obtained a similar projection vector for V_ORT by subtracting V_UCM–m,t,i from \islant-20%\partialθ_m,t,i. The Jacobian is described by an n × d = 2 × 8 matrix, where n represents the two dimensions of the task variable or the hammer's x-, y-positions and d represents the eight dimensions of the space of effector-level elemental variables, that is, corresponding to the eight DoFs. Accordingly, UCM had d − n = 8 − 2 = 6 dimensions and ORT had n = 2 dimensions. We calculated V_UCM and V_ORT for the mth monkey at the tth time slice averaged across all six strikes. Finally, to facilitate comparison between V_UCM and V_ORT, we normalized the magnitudes of the two projection vectors by the dimension of the respective subspaces:

and

In addition to estimating V_UCM and V_ORT across all eight DoFs (i.e. the whole kinematic chain), we were interested in the geometrical properties of the UCM. We thus estimated the projections of individual DoFs onto the UCM. Thus, in contrast to the standard UCM procedure, in which the projected deviations from the average vector are squared and summed across all elemental variables [24], we retained the squared, individual joint deviations, each of which represented the accumulated variability for the respective DoF in either V_UCM or V_ORT. We then calculated (i) V_UCM per DoF, V_ORT per DoF and R_V = V_UCM/V_ORT across the whole body, and (ii) V_UCM and V_ORT for each DoF in the lifting and lowering phases by averaging the values over time slices 1–12 and 13–24, respectively.

(f). Statistical analysis

We used linear mixed-effects models and constructed separate models for each response variable (table 3). We considered the fixed effects of body mass, hammer mass, strike phase, subspace, body part and DoF, whenever we incorporated these variables in the model; we dummy coded for strike phase, subspace and DoF. We accounted for inter-individual differences in each response variable by introducing a random effect of subject identity. Given the relatively small number of subjects (n = 5), we allowed only the intercept of this random effect to vary among individual monkeys. We performed all statistical analysis using MatLab and considered all statistical outcomes significant at the alpha level of 0.05.

Table 3.

Outcomes of linear mixed-effects models.

response variable	effect	estimate ± s.e.m.	t	d.f.	p	95% CI [lower, upper]
s.d. of hammer's x-position	body mass	0.006 ± 0.004	1.432	11	0.180	−0.003, 0.016
	hammer mass	0.019 ± 0.008	2.474	11	0.031*	0.002, 0.036
s.d. of hammer's y-position	body mass	0.011 ± 0.005	2.353	11	0.038*	0.001, 0.022
	hammer mass	−0.005 ± 0.005	−0.935	11	0.370	−0.016, 0.007
variability per DoF	body mass	−0.006 ± 0.016	0.392	51	0.697	−0.026, 0.038
	hammer mass	−0.061 ± 0.041	−1.513	51	0.136	−0.143, 0.020
	strike phase (lowering − lifting)	0.024 ± 0.029	0.772	51	0.444	−0.036, 0.081
	subspace (ORT − UCM)	−0.238 ± 0.029	−8.159	51	<0.001***	−0.297, −0.180
strength of synergy	body mass	0.000 ± 0.233	−0.001	51	1.000	−0.480, 0.480
	hammer mass	−2.229 ± 0.599	−3.724	51	0.001**	−3.465, −0.994
	strike phase (lowering − lifting)	−0.117 ± 0.430	−0.273	51	0.788	−1.004, 0.770
variability in individual DoFs	body mass	−0.003 ± 0.006	−0.515	429	0.606	−0.015, 0.009
	hammer mass	−0.034 ± 0.015	−2.226	429	0.027	−0.064, −0.004
	strike phase (lowering − lifting)	0.012 ± 0.011	1.133	429	0.258	−0.009, 0.034
	subspace (ORT − UCM) × DoF (θshank − θfoot)	−0.058 ± 0.044	−1.319	429	0.188	−0.143, 0.028
	subspace (ORT − UCM) × DoF (θthigh − θfoot)	−0.121 ± 0.044	−2.771	429	0.006**	−0.207, −0.035
	subspace (ORT − UCM) × DoF (θpelvis − θfoot)	−0.030 ± 0.044	−0.687	429	0.493	−0.116, 0.056
	subspace (ORT − UCM) × DoF (θtrunk − θfoot)	0.027 ± 0.044	0.623	429	0.534	−0.059, 0.113
	subspace (ORT − UCM) × DoF (θupper arm − θfoot)	−0.017 ± 0.044	−0.394	429	0.694	−0.103, 0.069
	subspace (ORT − UCM) × DoF (θlower arm − θfoot)	−0.123 ± 0.044	−2.808	429	0.005**	−0.208, −0.037
	subspace (ORT − UCM) × DoF (θhand − θfoot)	−0.514 ± 0.044	−11.761	429	<0.001***	−0.599, −0.428

*p < 0.05, **p < 0.005, ***p < 0.001.

3. Results

The monkeys produced hammer trajectories with highly repeatable spatial profiles across strikes (figure 4). Strike-to-strike variability in the hammer trajectory along the horizontal axis increased with hammer mass (t₁₁ = 2.474, p = 0.031, 95% CI [0.002, 0.036]; figure 4; table 3), and variability along the vertical axis increased with body mass (t₁₁ = 2.353, p = 0.038, 95% CI [0.001, 0.022]; figure 4; table 3). Thus, although using a heavier hammer was relatively more challenging, a heavier monkey could more flexibly alter the strike's amplitude (and, consequently, the kinetic energy at impact) independent of the hammer movement along the horizontal axis (electronic supplementary material, movies S1–S3).

Figure 4.

The monkeys produced hammer trajectories that were highly repeatable across hammers of different masses. (a) Hammer trajectory for Presente (body mass: 2.2 kg). (b) Hammer trajectory for Mansinho (body mass: 4.3 kg). The hammer was at the zenith at 50% movement. Shadings represent s.d.

Next, we examined the structure of motor variability. The monkeys employed strong joint synergies while lifting and lowering hammers, as V_UCM per DoF was considerably greater than V_ORT per DoF (t₅₁ = −8.159, p < 0.001, 95% CI [−0.297, −0.179]; figure 5a; table 3). The strengths of the synergies reduced with hammer mass (t₂₄ = −3.724, p < 0.001, 95% CI [−3.465, −0.994]; figure 5b; table 3), confirming that the task became increasingly challenging for the monkeys while using a heavier hammer.

Figure 5.

The monkeys employed strong joint synergies. (a) V_UCM and V_ORT per DoF. (b) The strengths of the synergies. Error bars indicate the s.e.m. (n = 5).

We examined the patterning of V_UCM and V_ORT across the eight DoFs. Given the direct relationship between foot stiffness and bipedal load-carrying capacity in humans [32], and the load-bearing role of the trunk while carrying loads bipedally in bearded capuchin monkeys [15,22], we anticipated that the monkeys would control movement more stringently in the feet and legs, and the torso (pelvis and trunk), compared with the hands and arms. Variability (V_UCM and V_ORT) decreased with hammer mass (t₁₁ = −2.226, p = 0.027, 95% CI [−0.064, −0.004]; figure 6; table 3). Although V_UCM was greater than V_ORT across all DoFs, the difference between V_UCM and V_ORT was significantly greater in θ_thigh (t₄₂₉ = −2.771, p = 0.006, 95% CI [−0.207, −0.035, 0.01]), θ_{lower arm} (t₄₂₉ = −2.808, p = 0.005, 95% CI [− 0.208,−0.037]) and θ_hand (t₄₂₉ = −11.761, p < 0.001, 95% CI [−0.599, −0.428]) than in θ_foot (figure 6; table 3). These results demonstrate that the monkeys stringently controlled the motion of the foot. They controlled the motion of the lower arm and hand rather loosely, showing a greater variability across strikes, although producing strikes with highly repeatable spatial profiles. Overall, monkeys use strong joint synergies to stabilize the hammer trajectory while maintaining bipedal stance.

Figure 6.

V_UCM and V_ORT for each DoF. (a) θ_foot. (b) θ_shank. (c) θ_thigh. (d) θ_pelvis. (e) θ_trunk. (f) θ_{upper arm}. (g) θ_{lower arm}. (h) θ_hand. Error bars indicate the s.e.m. (n = 5).

4. Discussion

In the present study, we identified the motor strategies bearded capuchin monkeys use to stabilize the hammer trajectory while cracking nuts with massive hammers. We hypothesized that if the requirement of maintaining a bipedal stance poses significant biomechanical and postural challenges, then to stabilize the hammer trajectory, capuchin monkeys would control motion in the lower body joints more stringently than motion in the upper body joints. To test this hypothesis, we analysed the kinematics of striking movements performed by five wild adult monkeys. We found that the monkeys produce hammer trajectories with highly repeatable spatial profiles. Using an uncontrolled manifold analysis, we show that the monkeys used strong joint synergies to stabilize the hammer trajectory while lifting and lowering heavy hammers. The monkeys stringently controlled the motion of the foot. They controlled the motion of the elbow and hand rather loosely, showing a greater variability across strikes.

Although it appears to the casual observer that lifting and lowering a massive hammer is challenging to the coordination of the upper limbs, variability in the upper body joints was not crucial in controlling the hammer trajectory. Instead, the challenge of maintaining a stable bipedal stance dictates the structure of motor variability in capuchin monkeys cracking nuts. Variability in the controlled subspace (which did not influence the hammer trajectory) was predominantly concentrated in the DoFs of the upper body, whereas variability in the controlled subspace (which influences the hammer trajectory) was predominantly concentrated in the DoFs of the lower body. In other words, the hammer trajectory was highly sensitive to variability in the motion of the foot, and only to a limited extent to variability in the motion of the arms and the hand. No such distinction was apparent in the trunk and the pelvis, as comparable magnitudes of variabilities in both controlled and uncontrolled subspaces characterized the motion of both these joints.

Here, our interpretation of the projections of individual DoFs onto the UCM surpasses the traditional interpretation that V_UCM and V_ORT for individual DoFs in the model reflect the geometry of the UCM [23–25]. Given that the UCM analysis is an analysis of covariation among elemental variables, the geometry of the UCM for the hammer trajectory defines the magnitudes of V_UCM and V_ORT for individual DoFs. For a multi-joint action such as that performed by the monkeys while cracking nuts, the angles between the UCM and individual axes corresponding to each DoF probably differ for different DOFs. This is because the geometry of the UCM is probably influenced by the distinct anatomical and physiological constraints on movements about each DoF. Accordingly, the magnitudes of the projections of individual DoFs onto the UCM reflect the underlying patterns of joint coordination. This possibility opens a new direction for the UCM analysis.

The challenges of balancing a massive hammer while standing bipedally generalize to other dynamic bipedal activities, including walking bipedally with or without a load, as accomplishing each task benefits from minimizing changes in the body's COM. Change in motor control of the lower limbs that resulted in a progressive reduction in the displacement of the body's COM are posited to have accompanied the evolution of obligate bipedalism [3–6]. Humans show strong joint synergies while walking that reduce variability in the body's COM [33], but Japanese macaques (Macaca fuscata) trained to walk bipedally show weaker joint synergies [34]. This discrepancy indicates that evolutionary changes in the structure of motor variability that served the postural demands of moving objects, such as lifting and lowering stones or other heavy objects, could also have supported occasional bipedalism. Thus, while capuchin monkeys do not represent a progressive step in the evolution of bipedalism in ancestral hominins, they highlight the potential involvement of varied activities, such as using percussive tools in bipedal stance, in the evolutionary transition from occasional to obligate bipedalism.

The present findings highlight that a specific structure of motor variability is required for members of a quadrupedal species such as the bearded capuchin monkey to lift and lower massive hammers. Capuchin monkeys must control motion in the joints of the lower body joints more stringently than motion in the upper body joints. Similar task demands probably demand similar motor strategies. Given the role of constraints in the development of coordination [35,36], phylogenetically distant species might show biomechanically comparable behaviour under identical constraints. We thus anticipate that other non-human primates that use stone hammers in bipedal stance will show similar motor strategies to lift and lower stone hammers.

Young capuchin monkeys practise striking nuts with stone hammers for three or more years before becoming proficient at cracking them [37,38]. Assuming that a key outcome of their motor learning is the stabilization of the hammer trajectory, (i) a larger reduction in V_ORT compared with V_UCM can occur, resulting in the emergence or strengthening of a synergy. (ii) Comparable reductions in both V_UCM and V_ORT can occur, resulting in an invariant strength of the synergy, R_V. (iii) Finally, a larger reduction in V_UCM compared to V_ORT can occur, resulting in a reduction in the strength of the synergy [23]. Given that each of the three scenarios is possible at different stages of motor learning [39,40], developmental changes in motor variability in young monkeys practising cracking nuts with stone hammers can reveal whether these monkeys discover joint synergies specific to this behaviour or, as we predict, refine existing joint synergies engaged in quadrupedal locomotion. Analogously, human infants might refine existing joint synergies engaged in quadrupedal locomotion (crawling) while mastering walking bipedally and carrying a load bipedally [41].

Ethics

The Institutional Animal Care and Use Committee (A2013 03-001-Y3-A2) at the University of Georgia (Athens, GA, USA) approved the present study.

Permission to carry out fieldwork

The Brazilian National Council for Scientific and Technological Development (CNPq, 002547/2011-2) and Authorization and Information System of Biodiversity (SisBio, 28689-5) permitted us to conduct research in Brazil.

Data accessibility

All data generated or analysed during this study are included in this published article (and its electronic supplementary material files).

Authors' contributions

M.M. and D.M.F. conceptualized the study; M.M. and D.M.F. conducted the experiment; M.M. and R.R. analysed and interpreted the data; M.M., R.R. and D.M.F. critiqued and/or revised the manuscript; D.M.F. provided the resources. All three authors approved the final version of the manuscript and agree to be held accountable for the content therein.

Competing interests

The authors have no competing interests to declare.

Funding

The University of Georgia (Athens, GA, USA) and the National Geographic Society (WW-051R-17) funded the present study.