Effort matching between arms depends on relative limb geometry and personal control

Abstract

Proprioception encompasses our sense of position and movement of our limbs, as well as the effort with which we engage in voluntary actions. Historically, sense of effort has been linked to centrally generated signals that elicit voluntary movements. We were interested in determining the effect of differences in limb geometry and personal control on sense of effort. In experiment 1, subjects exerted either extension or flexion torques to resist a torque applied by a robot exoskeleton to their reference elbow. They attempted to match this torque by exerting an equal effort torque (in a congruent direction with the reference arm) with their opposite (matching) arm in different limb positions (±15°). Subjects produced greater matching torque when their matching arm exerted effort toward the mirrored position of the reference (e.g., reference/matching arms at 90°/105° elbow flexion) vs. away (e.g., 90°/75° flexion). In experiment 2, a larger angular difference between arms (30°) resulted in a larger discrepancy in matched torques. Furthermore, in both experiments 1 and 2, subjects tended to overestimate the reference arm torque. This motivated a third experiment to determine whether providing more personal control might influence perceived effort and reduce the overestimation of the reference torques that we observed (experiments 3a and 3b). Overestimation of the matched torques decreased significantly when subjects self-selected the reference torque that they were matching. Collectively, our data suggest that perceived effort between arms can be influenced by signals relating to the relative geometry of the limbs and the personal control of motor output during action.

NEW & NOTEWORTHY This work highlights how limb geometry influences our sense of effort during voluntary motor actions. It also suggests that loss of personal control during motor actions leads to an increase in perceived effort.

INTRODUCTION

Proprioception involves perceptual processes related to the position, motion, and force generation of our body and limbs. It can be divided into three distinct percepts: position sense, kinesthesia (sense of motion), and sense of effort (sense of the muscular effort required to elicit forces) (Bastian 1888; Goldscheider 1898; McCloskey 1978; Proske and Gandevia 2012; Sherrington 1907). Position sense has been the most extensively studied of these sensory processes. Many studies have highlighted how muscle spindle afferents provide important information for perceiving the geometry of our limbs (Goodwin et al. 1972; Matthews 1982; Mileusnic et al. 2006; Scott and Loeb 1994). Cutaneous afferents also play an important role in position sense, particularly for receptors located in the hand (Collins et al. 2005; Edin and Johansson 1995; Moberg 1983).

Position sense, however, involves more than just sensory feedback from muscle and cutaneous afferents. Internal motor signals that are involved in generating movement contribute to position sense (Gandevia et al. 2006). These signals likely arise from efference copies of motor commands and are used to estimate limb position (Helmholtz 1867; Sperry 1950; Von Holst and Mittelstaedt 1950). As well, studies have suggested that position sense can be influenced by factors related to force generation. For example, fatiguing a limb segment (to modify the effort required to hold it in place) and increasing the isometric force applied by a limb segment have both been shown to impact our perception of its position (Rymer and D’Almeida 1980; Walsh et al. 2004; Watson et al. 1984; Winter et al. 2005). Additionally, isometric contraction (or willed contraction of experimentally paralyzed limbs) results in perceived limb displacement in the direction of applied effort (Gandevia et al. 2006; Rymer and D’Almeida 1980; Smith et al. 2009; Walsh et al. 2009). This relationship has been clearly shown at the finger and wrist (Gandevia et al. 2006; Rymer and D’Almeida 1980; Smith et al. 2009; Walsh et al. 2009), but the effect is somewhat less prominent at the elbow (Walsh et al. 2013). The above-described findings imply that a relationship between position sense and sense of effort exists, but this has been much less explored (Gandevia et al. 2006; Smith et al. 2009).

In general, sense of effort has been studied less extensively than position sense. However, efferent signals relating to the magnitude of the motor commands sent to our muscles appear to play an important role (de Morree et al. 2012; Gandevia and McCloskey 1977; Jones and Hunter 1983; Walsh et al. 2010, 2013). This is supported by studies in which subjects appear to match efforts according to the maximal strength of their arms (Brooks et al. 2013; Carson et al. 2002; Luu et al. 2011). Peripheral sensors [Golgi tendon organs (GTOs), muscle spindles, and skin receptors] have also been proposed to be involved in sense of effort (Brooks et al. 2013; Fleury et al. 1995; Lafargue et al. 2003; Luu et al. 2011; Sanes and Shadmehr 1995) but have been considered by many to be less important than centrally generated information (Carson et al. 2002; de Morree et al. 2012; Jones 2003; Jones and Hunter 1982; Proske et al. 2004; Scotland et al. 2014; Zénon et al. 2015).

In the present article, we address how specific peripheral and central factors influence our sense of effort using a bimanual effort matching paradigm in which subjects were instructed to match the effort generated by one arm to oppose externally applied loads with their other arm. Given that one’s sense of effort impacts the perception of limb position (Gandevia et al. 2006; Rymer and D’Almeida 1980; Smith et al. 2009; Walsh et al. 2009), we sought to explore the impact of altering limb geometry on sense of effort. We found a systematic overestimation in our initial effort matching experiments, even when limb geometry was identical for the two arms. Thus we performed a further experiment that demonstrated subjects were better at matching load levels if they selected the size of the applied load rather than it being controlled experimentally.

MATERIALS AND METHODS

This study was approved by the ethics board at the University of Calgary. Subjects provided written informed consent. A total of 35 subjects between 18 and 30 yr of age participated in the experiments. Inclusion criteria were normal or corrected-to-normal vision and self-reported right-handedness (for experiments 1a and 1b we verified this with a handedness score of >40, Modified Edinburgh Handedness Inventory; Oldfield, 1971). Before testing, subjects were excluded from the study if they had any self-reported neurological conditions (e.g., concussion, stroke, multiple sclerosis) or musculoskeletal conditions of the upper limb, examined via questionnaire. Subjects were asked to avoid strenuous physical activity for a day before testing to prevent any muscle soreness or fatigue that might impact performance.

Experimental Apparatus

Subjects performed experiments 1–3 in the KINARM Exoskeleton robot (BKIN Technologies; Scott 1999; Fig. 1A). The robot supported subjects’ arms in the horizontal plane (~85° shoulder abduction) using troughs that were sized to subjects’ arms. Limb segment lengths were also adjusted for each participant. This permitted horizontal extension and flexion motions at the shoulder and elbow (Fig. 1A). Additionally, the robot could apply torques to subjects’ shoulder and/or elbow joints. We calibrated the robot for each subject, such that a virtual reality system projected spatial targets and the position of the subject’s index fingertip (cursor) onto the workspace, and a metal shutter prevented direct vision of the arms (Fig. 1A).

Fig. 1.

Depiction of methodological apparatus and task structure. A: diagram of the KINARM exoskeleton robot, used for experiments 1a, 2, and 3. Subjects’ arms rested in fitted hand, forearm, and upper arm troughs. Subjects looked down on the task display. Motors pictured above the subjects’ shoulders could produce torques at the subject’s elbow and shoulder joints. B: Biodex dynamometer (experiment 1b) used to test maximum strength. Subjects pushed or pulled against the handle, shown as set up for extension; the handle was flipped for flexion to allow subjects to press against the frame (pictured in black) while their upper arm rested on a support. C–E: representations of the robotic experiments. Diagrams are not shown to scale. C: experiment 1. The robot applied a torque (either extension or flexion) to the reference arm (90°, shown as the left arm), which the subject resisted by producing an opposing torque to stay in the target (light gray arrow, dotted lines). Subjects matched the effort required to resist this torque (and stay in the circular target) with their matching arm (shown as the right arm) at one of the 3 pictured angles (75°, 90°, 105°) by exerting torque congruent in direction to that applied to the reference arm (indicated by an arrow that appeared on screen during the trial, shown in the diagram as a black arrow). The light gray arrows indicate the torques that subjects produced against the robot at each of the angles (shown for flexion condition only). These were not displayed on the screen during the task but are shown to illustrate conditions where, when matching, subjects either exerted torques “toward” the mirrored position of the reference arm, in a similar position, or “away” from it. Each condition was tested separately (right extension, right flexion, left extension, left flexion). D: experiment 2 was the same as experiment 1, except the angles of the reference arm were manipulated (75°, 105°) and only the right arm was used as the matching arm. E: experiment 3 was the same as experiment 1, except instead of a reference target, there was an arrow (so arrows appeared over both arms). The subjects applied torque in the direction of the arrow either 1) until it turned green or 2) until they were happy with the reference torque magnitude. Only the left arm was used as the matching arm. F: depiction of the torques for a single trial in experiment 1. The dashed line indicates the reference torque applied to subjects’ reference arm once their hand was in the reference target. The box indicates the 300-ms window, which was from 350 to 50 ms before the experimenter button press. Data were extracted from this window for analysis.

Experimental Protocol

To determine sense of effort, all robotic experiments involved contralateral matching tasks in which a reference torque was generated about the elbow of the reference arm. Subjects were then asked to match the effort they exerted in their reference arm by producing a matching torque with their opposite, matching elbow against a stiff spring (1 N·m·deg⁻¹) generated by the robot. Sense of effort was operationally defined as subjects’ ability to match the effort they felt in the reference arm (using their matching arm). The torques applied to subjects’ arms were always congruent in nature (i.e., both arms resisted extension or both arms resisted flexion). All subjects indicated they were not subjectively fatigued while performing the task. Data were sampled at 1 kHz.

Experiment 1a (robot): Robot-initiated reference torques.

Twenty-five subjects (14 women, mean age: 23 yr, range: 19–30 yr) participated in experiment 1a. At the start of each trial, subjects actively moved a cursor that represented the tip of the index finger of their reference arm (white, 0.5-cm radius) into a green “reference” target (0.75-cm radius) positioned at 30° horizontal shoulder flexion and 90° elbow flexion (Fig. 1C). After the subjects placed their reference arm, the robot applied an extension or flexion torque (2, 3, or 4 N·m) about the subjects’ elbow while maintaining their shoulder at 30° horizontal flexion (Fig. 1C). Subjects were instructed to resist the torque and keep the cursor aligned with the target. If subjects left the reference target, the target turned red.

While resisting the reference torque, subjects actively moved their other (matching) arm to a “matching” target, located at a 30° shoulder angle and either 75°, 90°, or 105° elbow flexion. Once subjects placed their matching arm in the target, their matching arm fingertip cursor and the matching target disappeared. A yellow arrow was presented on the side of the matching arm (Fig. 1C), which instructed the subjects to generate flexion or extension torque about their elbow, in accordance with the torque they generated from their reference arm. The torques in each arm were always congruent (i.e., extension in both arms or flexion in both arms). Elbow motion was resisted by a 1 N·m·deg⁻¹ spring. Subjects were instructed to move in the direction of the arrow and stop when they felt they matched the effort exerted by their reference arm. The trial ended when subjects verbally expressed a match in effort across the arms (the experimenter pressed a button). Subjects could take up to 30s before beginning the next trial. If subjects exceeded this, their trial was ended and discarded. Before actively moving their arms to the target to commence the next trial, subjects were given the option of freely moving their arms between trials.

One caveat to the methods described above is that, because of the spring used for the matching arm, the matching arm was not initially positioned at precisely 75°, 90°, or 105°. Depending on the torque being matched, the arm was initially placed either 2°, 3°, or 4° away from these angles to ensure that a perfect match (against a spring of a 1 N·m·deg⁻¹) would be made at angles of 75°, 90°, or 105°.

Our rationale for using the various matching elbow angles was to ensure that simply matching position could not be used as a proxy to match effort. We were particularly interested in this, because the eventual goal of this task was to explore sense of effort in individuals poststroke. However, investigating the different elbow angles allowed us to isolate conditions where the matching arm was exerting torque in a direction “toward” the mirrored position of the reference arm, in a similar position, or “away” from the mirrored position of the reference arm (Fig. 1C).

Subjects performed trials with both arms acting as the matching arm (right and left, R/L), in extension (both arms) and flexion (both arms). Thus there were four conditions (designated by matching arm/direction: right extension, right flexion, left extension, left flexion), each of which was tested separately in randomized order. Within each condition, subjects resisted three different torque levels with their reference arm (2, 3, 4 N·m) and matched this effort in three different matching arm configurations (75°, 90°, 105°), for a total of nine trial types per condition. Each trial type was tested six times (54 trials/condition). Within each condition, trials were shuffled pseudorandomly so that no two identical locations or torques were experienced twice in a row.

Experiment 1b (dynamometer): Maximum torque-angle relationship.

Previous studies have suggested that subjects use the proportion of their maximal strength exerted to gauge effort rather than the absolute forces elicited in the task (Adamo et al. 2012; Carson et al. 2002; Jones 2003; Jones and Hunter 1982; Scotland et al. 2014). In experiment 1b, we wanted to investigate whether responses from experiment 1a related to subjects’ maximal strength at each of the respective arm configurations. Ten subjects (5 women, mean age: 22 yr, range: 18–24 yr) from experiment 1a participated in experiment 1b. We used a Biodex dynamometer (Shirley, NY) to test subjects’ maximal elbow strength in the horizontal plane for each condition (arm/direction: right extension, right flexion, left extension, left flexion) and test angle (75°, 90°, 105°) from experiment 1a. Subjects rested their upper arms on a padded support (Fig. 1B) with the centers of rotation of the elbow and dynamometer aligned. The subject grasped the handle with their forearms pronated. During the trials, the dynamometer was locked into position to allow the subject to generate an isometric contraction. Subjects then generated a maximal voluntary contraction (MVC) by pressing the side of their hand against the handle attachment (Fig. 1B). Elbow joint angles were measured using templates corresponding to 75°, 90°, and 105° (while the subject performed a contraction) and checked with a goniometer. MVCs for each arm and angle were blocked together for testing and randomized sequentially by arm (R/L: subjects performed MVCs one arm at a time), direction tested (extension/flexion: for each arm, MVCs were performed for extension and then flexion, or vice versa), and elbow angle (75°, 90°, 105°), for a total of 12 unique trial types. Subjects performed three consecutive 3-s MVCs for each trial type with a minimum of 2 min of rest after each MVC (for a total of 36 trials). After the three test contractions were performed at all three angles (75°, 90°, 105°) for a given condition, subjects performed a reference contraction at the same angle they started with to determine whether their elbow muscles were fatigued during the MVC trials (36 trials + 4 fatigue reference contractions = final total of 40 trials). We took the maximum torque from each 3-s MVC and compared these values among the three trials at each respective elbow angle, taking the highest value as the subjects’ maximum voluntary strength at that angle. If the fatigue reference was the strongest at that angle, we used that value (10 of 40 occasions).

Experiment 2 (robot): Robot-initiated reference torques with larger angular differences between arms.

Experiment 2 was conducted to evaluate matching behavior when the arms were farther apart compared with experiment 1. Ten subjects (6 women, mean age: 24 yr, range: 20–27 yr) participated in experiment 2. Methods for experiment 2 were the same as for experiment 1, except that in the robotic task, instead of the reference arm being at 90°, it was tested at angles of 75° and 105° (Fig. 1D), and only the right arm served as the matching arm. The matching arm (left arm) was tested at the same three angles (75°, 90°, 105°) and same three torques (2, 3, 4 N·m). The order of conditions was randomized, and the 18 unique trials were pseudorandomized such that the same trial was never repeated twice in a row.

Experiment 3a (robot): Subject-initiated reference torques.

Experiment 3a was performed to investigate whether subjects would be more accurate at matching the reference torque when it was self-initiated (i.e., rather than the robot producing the reference torque and requiring the subject to resist as in experiment 1a, subjects voluntarily initiated the reference torque against the robot up to a set point). It also allowed us to remove the influence of a specific visuospatial goal (the target) on the reference arm. The same 10 subjects from experiment 2 participated in experiment 3a. Due to time constraints, we only tested the left arm as the matching arm.

Experiment 3a was the same as experiment 1a, except that subjects initiated the reference torque themselves against a 1 N·m·deg⁻¹ spring. This required us to make a few changes to the experimental paradigm. At the beginning of a trial, subjects moved to both reference and matching targets before any robotic spring loads were elicited. The subjects’ reference fingertip cursor then disappeared, and a red arrow appeared over their reference arm (Fig. 1E). Subjects extended/flexed their reference arm in the direction of the arrow against a 1 N·m·deg⁻¹ spring. The arrow turned green (at a reference elbow angle of 90°) to indicate the subject reached the target reference torque. The arrow remained green if the measured torque from the robot was within ±0.25 N·m of the target torque (2, 3, 4 N·m). The arrow turned red when the subject did not produce enough torque and disappeared when the subject produced too much torque. Once the target torque was reached with the reference arm, a yellow arrow appeared on side of the matching arm, and the subject exerted torque with their matching arm to match the effort produced in the reference arm (as in experiment 1a).

Experiment 3b (robot): Robot-initiated torques vs. subject-initiated reference torques with magnitude under personal control.

In experiment 3b, 8 subjects [5 women, mean age: 26 yr, range: 21–28 yr; 4 of whom (2 women) also performed experiments 2 and 3a] performed two task variants: a robot-initiated reference torque task, similar to experiment 1a (“original”), as well as a second task in which subjects initiated their own torques but were not restricted to a specific reference torque (“free choice”) and determined the magnitude of the reference torque themselves [the same task as experiment 3a (Fig. 1E), except the arrow remained green the entire time, allowing the subject to have personal control over the reference torque magnitude].

Due to time constraints, we only tested flexion with the left arm as the matching arm, and it was also the condition with the most noticeable changes from experiment 3a. The original task variant had set robot-initiated reference torques: 2 (light) and 4 N·m (heavy). To keep the torque magnitudes in a somewhat similar range in the free choice task, we familiarized the subjects with five practice trials using the magnitudes of a light/heavy (2/4 N·m) torque before executing the experiment. In these practice trials, the experimenter verbally confirmed whether they achieved light (2 N·m) or heavy (4 N·m) ±0.25 N·m torque with their reference arm only. No feedback was provided for the free choice task itself.

During experiment 3b, subjects matched each torque (light/heavy) twice at each elbow angle (75°, 90°, 105°) for a total of 12 trials in the original variant (where 2/4 N·m torques were pseudorandomized). In the free choice variant, subjects were instructed to produce what they felt was a light torque or a heavy torque at each of the above angles in pseudorandom order. We repeated these tasks three times to make up three testing blocks (each individual trial was tested 6 times). The blocks were randomized to begin with either the subject-initiated free choice condition or the robot-initiated original task.

Data Analysis

Data were analyzed in MATLAB (The MathWorks, Natick, MA) on a desktop computer. All trials were examined for the timing of the logged event where the subject confirmed a match. Subject matching performance was based on the robot torque levels that resisted each limb during a window from 350 to 50 ms before the button press by the operator (Fig. 1F). Subjects were encouraged to hold their matching torque until they felt it ramp down, so we are confident that we captured their matching torque within this time window. Trials were excluded from analysis if the torque produced by the subject during the 300-ms period varied by more than 0.5 N·m (<1% of trials). Statistical data analysis was done using IBM SPSS 23 (Chicago, IL) with alpha levels set to 0.05.

Experiment 1a.

We began with a four-way (2 × 2 × 3 × 3) repeated-measures analysis of variance (RM ANOVA) examining the effects of arm (R/L), direction (extension/flexion), reference torque (2, 3, 4 N·m), and angle (75°, 90°, 105°) on both matched torque and coefficient of variation (CV; a measure of within subject variability). Due to significant interactions between direction and angle, we pursued two three-way (2 × 3 × 3) RM ANOVAs separately for extension/flexion, examining the effects of arm (R/L), reference torque (2, 3, 4 N·m), and angle (75°, 90°, 105°). Also, due to the effects of angle (75°, 90°, 105°) and direction (extension/flexion) that we observed, we wanted to assess performance when both arms were in the same position. To do this, we used a three-way (2 × 2 × 3) RM ANOVA with the data for each arm at 90° to test the effect of arm (R/L), direction (extension/flexion), and reference torque (2, 3, 4 N·m) on the amplitude of the matched torque.

Experiment 1b.

We began with a three-way RM ANOVA to examine the effects of arm (R/L), direction (extension/flexion), and angle (75°, 90°, 105°) on subjects’ MVCs. Due to a significant interaction between direction and angle, we carried out two two-way (2 × 3) RM ANOVAs (for extension/flexion) examining the effects of arm (R/L) or angle (75°, 90°, 105°). These tests were performed on subjects’ raw torque values (N·m) obtained during their MVC.

Experiment 2.

Due to previously observed differences for extension and flexion in experiment 1, two three-way RM ANOVAs (for extension/flexion) were performed to examine the effects of reference torque (2, 3, 4 N·m), reference target (75°, 105°), and angle difference between arms (±30/ ± 15/0°) on matched torques. We then performed one-way RM ANOVAs at each reference torque (2, 3, 4 N·m)/target (75°, 105°) combination for extension and flexion to test for differences in matched torque (measured in absolute N·m). Two-tailed paired t-tests were performed at each reference torque to examine the effect of angle on matched torque when there were no relative differences in arm position (reference and matching arms were at the same angle, either 75° or 105°).

Experiment 3a.

Mixed-model RM ANOVAs (for extension/flexion) investigated the effects of experiment (between-subject factor: experiment 1a vs. experiment 3b) and the effects of reference torque (2, 3, 4 N·m) and angle (75°, 90°, 105°; within-subject factors) on matched torque.

Experiment 3b.

Because the reference torques set by subjects in the free choice variant were not always consistent, the outcome variable of interest for experiment 3 was the difference between the reference and matching torques. A three-way RM ANOVA [variables: torque (light/heavy), angle (75°, 90°, 105°), and reference torque generation (original/free choice)] was performed on the absolute differences (in N·m) as well as the percentage of the reference torque exerted with the matching arm.

Post hoc testing for all ANOVAs included pairwise t-tests, which were corrected using the Bonferroni correction for multiple comparisons. For RM ANOVAs, if Mauchly’s sphericity test was significant, we report significances using the Greenhouse-Geisser (Greenhouse and Geisser 1959) correction. Unless indicated otherwise, value are means ± SE.

RESULTS

Experiment 1: Influence of Arm Geometry on Sense of Effort

In experiment 1, subjects were able to differentiate the magnitude of each reference torque in their matches [as shown by a significant effect of reference torque in both extension (F_1.2,26.8 = 101.05, P < 0.001) and flexion (F_1.2,29.9 = 102.41, P < 0.001); Fig. 2]. Subjects overestimated the reference torques when matching. On average, subjects produced 133.7% ± 5.0% of the torque applied to their reference arm with their matching arm. When their reference and matching arms were at different joint angles, subjects matched with higher torque magnitudes when they produced torque toward the angle of their reference arm [i.e., in flexion, when the matching arm was positioned at 105°, such that it was pulling “toward” the mirrored angle of the reference (90°) to match; Figs. 1 and 2], rather than away from it [i.e., in flexion, when the matching arm was positioned 75°, such that it was pulling away from the mirrored angle of the reference (90°) to match; Figs. 1 and 2]. With flexor matching, the greatest overestimation was seen in the most extended matching arm position (105°; Fig. 2, A and B, extension: F_1.4,33.3 = 93.54; flexion: F_1.3,31.3 = 24.04 P < 0.001). Similarly, we also found that the within-subject variability of the effort estimate (measured using the CV) increased with the angular separation between arms (F_2,48 = 14.39/23.56 and P < 0.001 for extension/flexion, respectively).

Fig. 2.

A: matched torques produced by individual subjects. Individual subject data (n = 25) are shown for the 3 reference torques (top to bottom) at various matching elbow angles (x-axis) for experiment 1a. Thin lines show individual subjects’ mean torques for each angle, whereas thick lines represent the means across subjects. Dashed gray lines represent the magnitude of reference torques. B: mean (±SE) matching performance on experiment 1a (n = 25), separated by condition. Dashed lines represent reference torque. *P < 0.05, significant differences in pairwise comparisons (Bonferroni corrected) between subjects’ matched torques at each reference torque magnitude across angles (horizontal) and reference torques (vertical).

When we compared the data for matches in the same position (both arms at 90°), we found that subjects produced higher torques when they were matching a torque applied to their left arm (reference arm) with their right arm as the matching arm (F_1,24 = 4.89, P = 0.035; Fig. 2B) and when the they were flexing their elbows vs. extending them against the robot (F_1,24 = 9.72, P = 0.005, Fig. 2B). We observed that individual performance appeared stable over time. By observation, we did not detect differences between the torques elicited at a given torque angle combination based on the previous trial’s torque.

Experiment 1b: Maximum Strength at Robotic Test Angles

There were no significant differences between subjects’ right and left arm MVCs at each angle tested [overall means of 50.05 ± 5.98 and 52.31 ± 6.19 N·m for extension (F_1,9 = 3.03, P = 0.116) and 58.55 ± 7.02 and 57.58 ± 7.98 N·m for flexion (F_1,9 = 0.65, P = 0.441) for right and left arms, respectively]. Elbow extension MVCs were highest at 105° and lowest at 75° (F_2,18 = 23.910, P < 0.001), whereas MVCs in elbow flexion were similar across all angles (F_2,18 = 3.57, P = 0.089; Fig. 3).

Fig. 3.

Lines contrasting the mean (±SE) normalized maximal voluntary contraction (MVC) strength (in N·m) in experiment 1b and robotic performance on experiment 1a for the 10 subjects who participated in both. Normalization for both MVC and matched torque was achieved by taking a subject’s performance at each of the 3 individual angles and dividing it by the mean of all 3 values.

Figure 3 shows a comparison of subjects’ normalized MVC for each angle against robotic effort matching performance. Subjects’ MVCs did not parallel the trends for matching effort in the robot. For extension, subjects matched with the weakest torque where they showed the strongest MVC. For flexion, despite no significant differences in nonnormalized MVC across elbow positions (F_1.0,9.4 = 3.57, P = 0.089), robotic testing showed a significant effect of angle (Figs. 2 and 3).

In general, the reference torques we tested were small relative to subjects’ MVCs. The 2, 3 and 4 N·m reference torques were, on average, 3.9% MVC (range: 1.8–7.4% MVC), 5.8%MVC (range: 2.7–11.0% MVC), and 7.8% MVC (range: 3.5–14.6% MVC), respectively.

Experiment 2: Influence of Larger Differences in Arm Geometry on Sense of Effort

In experiment 2, we sought to determine whether the effects of arm geometry on sense of effort would be amplified when the difference in relative elbow angle was increased. We hypothesized that increasing the relative elbow angle would result in subjects applying even more or less matching torque, depending on whether effort was made in a direction that pointed toward or away from the mirrored angle of the reference arm. Indeed, the effect of relative arm geometry tended to increase when the discrepancy in angles between arms was larger (Fig. 4).

Fig. 4.

Results for experiment 2 (n = 10), examining greater discrepancies in position with reference targets at 75° and 105°. Mean matched torques are plotted for extension and flexion, arranged by relative angle differences. Starting at left, the plots begin at −30°, which denotes the arms being at elbow angles that are 30° different, with the matching arm exerting torque away from the reference, −15° indicates a similar scenario with 15° difference between arms, and “0” indicates that the matching arm is positioned at the same angle as the reference. The notation applies for the second reference target, except in these cases they are denoted as “+” because the matching arm exerted torque toward the position of the reference arm. Actual reference and matching arm positions are listed at bottom, including diagrams that show the position of the arms at each angle. Error bars indicate SE. *P < 0.05, significant effect of angle for the reference torque indicated by horizontal bars; n.s., not significant (Bonferroni corrected for the number of comparisons in each condition). Thick, dark gray bars indicate significance values for the ANOVA; thin, light gray bars underneath indicate significance for the post hoc pairwise comparisons.

Similar to experiment 1a, subjects produced the greatest matching torques when, to match, they produced torque that was oriented toward the mirrored angle of their reference arm, rather than away from it (Figs. 1 and 4). When subjects were matching with both arms in effectively the same position (indicated by 0° on the x-axis of Fig. 4), their perceived effort, measured by the matching torque exerted, was larger for the extension condition when both arms were at angles of 75° vs. 105° at 3 N·m (t = 3.46, P = 0.007), but not at 2 or 4 N·m (t = 1.14, P = 0.285 or t = 2.29, P = 0.048, respectively, Bonferroni corrected α = 0.016). For flexors, it was larger when both arms were at 105° vs. 75° for two loads (2 N·m: t = 5.95, P < 0.001; 4 N·m: t = 3.06, P = 0.013), but not at 3 N·m (t = 2.17, P = 0.058, Bonferroni corrected α = 0.016; Fig. 4).

When subjects’ arms were in different positions, their matching torques scaled depending on the relative elbow angles of their arms (extension: F_1.1,10.2 = 19.26, P = 0.001; flexion: F_2,18 = 8.13, P = 0.003; Fig. 4). In the extension condition, subjects exerted greater matching torques at the more extended reference angle (105°) than the more flexed reference angle (75°; F_1,9 = 22.58, P = 0.001; see Fig. 4). For the flexion condition, the less flexed reference angle of 105° showed, on average, significantly greater matched torques than 75° (F_1,9 = 5.99, P = 0.037). We have indicated the statistical results of the one-way ANOVA performed to examine for differences between angles at a given reference target in Fig. 4.

Experiment 3a: Robot-Initiated vs. Subject-Initiated Matching Torques

The goals of experiment 3a were to determine if subjects would be more accurate when they initiated their own reference torque (rather than it being initiated by the robot). In general, subjects displayed the same torque matching patterns as in experiment 1a: for both extension and flexion, we saw the same significant within-subject effects of reference torque and angle on matched torque as in experiment 1a (Fig. 5; P < 0.001), with no significant interaction between the matching torques from experiment 1a vs. experiment 3a or for reference torque level/angle.

Fig. 5.

Experiment 3a matching torques (self-generated reference version of experiment 1a) are plotted as solid lines that are shaded for each angle. Results from experiment 1a are shown for contrast as dotted lines (similarly shaded). Error bars indicate SE. *P < 0.05, significant effects of angle and reference torque across both experiments (Bonferroni corrected within conditions).

For extension, we found no significant difference between the mean matched torques for experiment 1a vs. experiment 2 (difference: 0.17 ± 0.24 N·m, F_1,33 = 0.48, P = 0.494; Fig. 5). For flexion, there was a nonsignificant trend for reduced matched torques in the self-initiated condition in experiment 1a vs. experiment 2 (difference: 0.54 ± 0.31 N·m, F_1,33 = 3.15, P = 0.085; Fig. 5).

Experiment 3b: Robot-Initiated vs. Subject-Initiated Free Choice Matching Torques

We designed experiment 3b to determine whether subjects would 1) be more precise and 2) show less variability in matching torques if they were given more personal control over the reference torque. The data from Fig. 6 suggest that when subjects were free to produce their own reference torque magnitude, they had less error. Instead of overestimating the reference torques by exerting 118.3% ± 2.2% as matching torque, on average, as subjects did in the original variant of experiment 3b, they exerted 95.6% ± 1.8% in the free choice variant of experiment 3b (F_1,7 = 7.916, P = 0.026; Fig. 6A). Thus, with the addition of personal control, the matching and reference arm torques were closer in magnitude (absolute difference; Fig. 5A). The descriptive data suggest that, in contrast to experiment 1 (where subjects overestimated their matching torques on average), in experiment 3b subjects underestimated the amount of torque needed to match the reference arm (expressed by a negative relative difference in torques across arms; Fig. 6A).

Fig. 6.

Experiment 3b comparisons of the robot-initiated reference torque (original) task version (dark gray) and the subject-initiated reference torque with the self-initiated (free choice) version (light gray). A: absolute difference (left) and relative difference (right) in torques between the mean matching-reference torques across all subjects for the original and free choice task versions. Absolute differences illustrate the positive values of the difference between matching minus reference arm torque; relative differences illustrate the difference (either positive or negative) between matching minus reference arm torques. Black lines connect individual subject means for each task version. B: mean matching vs. reference torques for each subject (circles) for the original (dark gray) and free choice (light gray) versions. Least-squares regression lines are also plotted for the original (dark gray) and free choice (light gray) task versions.

In addition to a significant difference between the mean matched torques for the original vs. free choice conditions (F_1,7 = 8.36, P = 0.027; Fig. 6), subjects tended to underestimate the reference torque more when they produced heavier, rather than lighter, torques with their reference arm (F_1,7 = 2.319, P = 0.026; Fig. 6B). The effect of matching arm angle found in experiments 1 and 2 persisted (F_2,14 = 4.199, P < 0.001; Fig. 6).

DISCUSSION

In this study, we investigated sense of effort using a contralateral elbow torque matching task performed in the horizontal plane. We examined various reference and matching arm positions and different methods of reference torque initiation (robot vs. subject vs. personal control). We also contrasted effort matching to subjects’ maximum force production in the elbow configurations tested in the robotic matching tasks. We report three key findings. First, differences in arm position influenced perceived effort (experiment 1a), in a way that was not directly linked to maximal strength (experiment 1b), and these effects appeared to be amplified when the discrepancy in elbow angles was increased (experiment 2). Second, we observed that when the arms were mirrored, matching torques varied based on the angle of the arms (experiment 2). Third, subjects tended to overestimate the applied torque when the reference torque was robot initiated (experiment 1 and experiment 3b original variant), and this was significantly reduced when subjects were given personal control over the reference torque they produced (subjects tended to underestimate their matches of the reference torque in experiment 3b free choice variant).

Influence of Angle Discrepancy on Perceived Effort

A common finding in all our experiments is that subjects tended to increase their matched effort when exerting torque toward the angle of the reference and to decrease it when exerting torque away from the angle of the reference. One possible explanation for this is that subjects generated similar drive to the muscles, but peripheral factors, notably the force-length relationship of muscle, may have altered the amount of force generated by the subject. However, when we examined this hypothesis by quantifying each subject’s MVC at different joint angles, we found no systematic relationship between variations in MVC and variations in perceived effort. In fact, subjects’ extensor MVC was greatest at the joint angle where their effort estimate was the smallest (Fig. 3). This is opposite to the effect that would have been expected if subjects matched proportionally to maximum strength. It is possible that, due to the low levels of torque employed in our experiments (2, 3 and 4 N·m), some of these differences might be associated with passive forces related to the length of the muscles. However, we also believe that our results suggest that subjects used peripheral information about limb geometry to help judge effort.

Consistent with the idea that relative arm position influenced perceived effort, when we increased the angular difference between the reference and matching limbs (experiment 2), subjects applied even larger matched torques in the direction of the reference joint angle and smaller matched torques away from the reference joint angle (Fig. 4). Given this finding, we suggest that afferent information about limb position played an important role in effort judgements. Recent research has suggested that, in addition to position and velocity, muscle spindles may also encode effort or force (Luu et al. 2011; Ting et al. 2018). Spindles are tuned internally by fusimotor neurons supplying the intrafusal muscles (Hulliger 1984), and fusimotor drive can be modified by the task (Prochazka et al. 1988). As such, spindles can operate as peripheral efference copy readers (Dimitriou and Edin 2010), increasing their discharge with stronger isometric contractions (Vallbo 1974). If subjects relied, even partially, on muscle spindle feedback in our task, this may have contributed to the differences observed at respective angles. Another potential reason for the influence of position, and thus spindle feedback, is the structure of our task. Subjects were pressing against extremely low torques (average of 6% MVC) while attempting to hold their arm in a specified position. This may have caused subjects to be more aware of the position of their reference arm (and the associated feedback from muscle spindles), which could have influenced their behavior. In summary, it is likely that afferent information from the periphery related to limb position contributed to sensing effort in our tasks.

Effect of Angle When Both Arms Were in the Same Position

To our knowledge, no previous study has examined effort matching across the workspace with both arms at the same position. When subjects’ agonist arm muscles were shorter (105° for extension torques and 75° for flexion torques), their matched torques tended to be lower compared with when their agonist muscles were longer. This is an intriguing finding. It is unlikely that it can be explained by differences in strength when in the limb was placed in different geometries, because matching and reference arms were in the same position and should have been similar in strength (verified by dynamometry for experiment 1b).

Another possible explanation for the matching pattern observed is that differences in acuity across the workspace exist for sense of effort, as has been reported for position sense (Fuentes and Bastian 2010; Wilson et al. 2010). In our experiment, however, the areas of the workspace in which subjects overestimated were opposite in flexion and extension (for flexion, subjects overestimated when their arm configuration was more extended, and for extension, subjects overestimated when their arm configuration was more flexed). This leads us to believe that the differences arise from muscular, not end point, factors.

Estimation of Torques and the Role of Personal Control

In our first experiments (experiments 1 and 2), subjects had no control over the applied torque on the reference arm. In experiment 3a, we had the subjects voluntarily initiate the torque rather than simply resist the robot as in experiments 1 and 2. However, the final level or torque was predetermined by the robotic task. We did this in part to test whether the overestimation observed in experiment 1a might be due to the phenomenon of “sensorimotor attenuation,” in which it is postulated that subjects predictively attenuate the resultant sensory feedback from their voluntary actions, due to observations that subjects overestimate when actively matching forces that they receive passively (Bays and Wolpert 2007; Blakemore et al. 1999; Shergill et al. 2003). We did not find a significant reduction in matching torques in experiment 3a (Fig. 5), and, in fact, subjects continued to overestimate the torque in their reference arms when matching. This may have been because our study was an active task, unlike other studies of sensorimotor attenuation (Blakemore et al. 1999; Shergill et al. 2003), and subjects were required to maintain contraction throughout.

Thus our data did not suggest that the overestimation observed when subjects were matching robot-initiated torques was due to sensorimotor attenuation. We hypothesized that there might be another factor involved. It is reasonably well established in the pain literature that mindfulness or establishing personal control of the sensation of pain can alter the perception of pain (Beck et al. 2017; Helmchen et al. 2006; Salomons et al. 2007; Wang et al. 2011). In experiment 3b, we tested whether having more personal control of the applied torque (the ability to self-initiate and choose the magnitude of the reference torque) would further decrease the overestimation of sense of effort seen in the earlier experiments (experiments 1 and 2). Ultimately, giving the subjects more control reduced the overestimation to the point where the matching torques produced were nearly undistinguishable from the reference torques (Fig. 6). This supports the idea that personal control is an important factor in the estimation of self-produced forces.

It has previously been shown that individuals are more accurate in estimating limb position when they actively place their limb, rather than it being placed passively (Fuentes and Bastian 2010; Gritsenko et al. 2007; Paillard and Brouchon 1968). In all these studies, the final configuration of the subjects’ arms was determined by the experimenter. This is similar to what we tested with experiment 3a, where the reference torque was set by the task program but subjects voluntarily initiated their own torque up to that point (rather than the robot applying reference torque that the subject resisted to remain in position). Experiment 3b was different, because in addition to subjects voluntarily initiating their reference torques, they also were given the ability to control the final magnitude. We found that adding this element of personal control became key in reducing subjects’ overestimation of the reference torque and enabling them to become more accurate.

Several other factors could have influenced overestimation, including the lack of visual feedback of the matching arm (overestimation may increase in situations without visual feedback; Therrien and Balasubramaniam 2010; Therrien et al. 2011, 2013), any co-contraction of the reference arm, and use of a spring load against which subjects pressed to elicit the matching torque (people are worse at differentiating spring stiffness than constant forces; Jones 1989; Jones and Hunter 1990; Roland and Ladegaard-Pedersen 1977).

Furthermore, the variations we observed in sense of effort raises the question, “how important is accuracy in sense of effort, and is it a problem that estimation is not always accurate?” Although effort estimation using our matching task was not perfectly accurate, we showed that it scaled with the torque magnitudes presented to the matching arm. On a single task, the ability to match efforts between limbs within a few newton meters may have relatively little value. However, in the “real world,” we would argue that sense of effort is important to accomplish day-to-day tasks. Sense of effort also likely relies on many forms of sensory feedback, such as vision, position sense, movement sense, and cutaneous sense, to accurately accomplish a task (for example, balancing two full bowls of soup on a tray or pouring a jug full of water). In the present work, it is even possible that these different forms of feedback influenced subject’s matches.

Perhaps it is only when other forms of proprioception are impaired that the inaccuracies of our sense of effort become important. Rothwell et al. (1982) reported a case of a man who acquired large-fiber sensory deafferentation who was able to drive his old car by sensing the effort required to press the gas pedal but could not learn to drive any other cars. In previous studies, it has been shown that when trained, deafferented subjects can reproduce similar levels of effort simply based on their outgoing commands, but that their reproduced effort is not as consistent over time as that of subjects with intact proprioception (Lafargue et al. 2003; Luu et al. 2011). We have observed similar sense-of-effort impairments in subjects with stroke, and developing a task that could identify these was our original goal.

Conclusion

The results from our experiments show that there is a relationship between perceived effort and 1) the relative angles of the elbow, 2) the position of the arms when both at the same angle, and 3) the level of personal control over torque generation. We show that personal control is important in torque estimation and accuracy. Based on the dependence of matching torque on the angles of the arms, it is likely that subjects used feedback information from peripheral sources related to limb geometry to gauge effort. Our work provides supporting evidence for the notion that proprioception, including effort, is multisensory and integrative.

GRANTS

The laboratory and support staff were supported by Canadian Institutes of Health Research Grant MOP 106662. L. M. Logan was supported by the National Sciences and Engineering Research Council Canada Graduate Scholarships-Master’s Program and an Alberta Innovates Graduate Studentship.

DISCLOSURES

S. H. Scott is the inventor of the KINARM exoskeleton and continues to serve as the Chief Scientific Officer for BKIN technologies in Kingston, Ontario, Canada.

AUTHOR CONTRIBUTIONS

L.M.L., J.A.S., T.C., S.H.S., and S.P.D. conceived and designed research; L.M.L. performed experiments; L.M.L. analyzed data; L.M.L., J.A.S., T.C., S.H.S., and S.P.D. interpreted results of experiments; L.M.L. prepared figures; L.M.L. drafted manuscript; L.M.L., J.A.S., T.C., S.H.S., and S.P.D. edited and revised manuscript; L.M.L., J.A.S., T.C., S.H.S., and S.P.D. approved final version of manuscript.