Kinematic characteristics of manual guidance across expertise levels in a motion reproduction task: implications for physical therapist education

Abstract

[Purpose] We aimed to quantitatively evaluate manual guidance skills based on motion observation and identify kinematic characteristics associated with expertise in physical therapists. [Participants and Methods] Participants included skilled physical therapists, novice physical therapists, and individuals without experience in movement instruction. After observing an obstacle-avoidance reaching movement, participants reproduced the movement by manually guiding a humanoid model. We recorded fingertip kinematics using a three-dimensional motion analysis system and calculated absolute errors relative to the reference motion for movement time, trajectory length, peak velocity, acceleration, timing, and normalized jerk costs. After excluding participants with missing coordinate data, ten participants per group were analyzed. [Results] The skilled physical therapist group showed fewer errors in movement time, peak velocity timing, and negative peak acceleration timing than the other groups. The trajectory length did not differ among groups, and the normalized jerk cost error showed no significant group differences. [Conclusion] We found that expertise-related differences in manual guidance appeared more clearly in temporal parameters than in spatial trajectory or waveform structure. These findings suggest that temporal aspects of movement may represent one fundamental component of manual guidance expertise and may contribute to the development of objective evaluation methods and educational approaches in physical therapy.

INTRODUCTION

Physical therapists frequently use manual guidance to assist and guide patient movements through hand contact with a patient’s body. Although this technique is conceptually related to body handling in approaches such as the Bobath concept, it is now applied to a wide range of clinical conditions. However, the terminology, implementation, and theoretical basis of manual guidance remain unclear and have not yet been standardized¹⁾.

Previous studies have analyzed manual guidance techniques quantitatively. Haarman et al.²⁾ quantified the direction, magnitude, and timing of forces during gait training, whereas other studies examined pelvic guidance and handling strategies during functional movements^{3, 4)}. These studies demonstrated that therapist assistance can be characterized using biomechanical parameters.

However, several methodological limitations remain. First, reference movements were not clearly defined, and manual guidance relied on the therapists’ internal representations of movement. In tasks with high degrees of freedom, such as gait or sit-to-stand movements, the variability in movement patterns makes it difficult to determine whether the recorded data reflect appropriate guidance. Second, the interaction between the therapist and participant could not be fully eliminated. Active participation by the participants and changes in passive joint resistance may have influenced the measured outcomes. Third, some studies analyzed only a single therapist, limiting the identification of shared technical characteristics. In addition, previous research has shown that guidance strategies vary according to therapist expertise⁵⁾.

These limitations suggest that therapist-specific abilities in reproducing target movements have not yet been sufficiently isolated. Examining the accuracy with which therapists reproduce observed movements under conditions that minimize interaction between the therapist and the participant may help clarify the fundamental components of manual guidance skills.

To address these issues, this study employed three methodological strategies. First, a humanoid model based on actual body dimensions was used to eliminate therapist–participant interactions while preserving anthropometric characteristics⁶⁾. Second, an obstacle-avoidance reaching task with constrained joint degrees of freedom was used to limit movement variability. Third, therapists with different levels of expertise were compared.

This study aimed to quantitatively evaluate manual guidance skills based on motion observation using a humanoid model and to identify the kinematic characteristics associated with the expertise of physical therapists.

PARTICIPANTS AND METHODS

This observational experiment compared the manual guidance performance of skilled physical therapists, novice physical therapists, and individuals without experience in movement instruction. The study included a preliminary phase to establish a reference motion, and a main experiment to evaluate movement reproduction (Fig. 1).

Fig. 1.

Study flow of the preparation phase and main experiments.

In the preparation phase, anthropometric measurements were obtained, reference motion data were recorded using three-dimensional motion capture, a humanoid model was constructed, and an observation video was created. In the main experiment, skilled physical therapists (sPT), novice physical therapists (nPT), and controls observed the video, completed practice trials, and reproduced the movement. Motion data were recorded and kinematic errors were calculated. Participants with missing data were excluded from the analysis.

All participants received written and verbal explanations of the study and provided written informed consent before participation. The study was approved by the Research Ethics Committee of Tohoku Bunka Gakuen University (approval no. 18-3).

In the preparation phase, three-dimensional kinematic and video data of a reference movement performed by a healthy young male were collected, and a humanoid model was constructed based on the anthropometric measurements of the same participant. The participant performed the obstacle-avoidance reaching task used in the main experiment (Fig. 2). To reduce the degrees of freedom of the upper limb, the participant’s wrist and finger joints were immobilized using a wrist protector (PeaceCafe Co., Japan, Kanagawa, Japan) and a metal plate. An infrared reflective marker was attached to the distal phalanx of the middle finger, and the participant was seated with the trunk secured to the backrest. The participant’s upper-limb length, upper-arm length, forearm length, hand length, thigh length, floor-to-acromion distance, and floor-to-fibular-head distance were measured using a Martin-type anthropometer. Based on these measurements, the target for the reaching movement (a circular disk, 0.30 m in diameter) and the vertical obstacle (an aluminum pipe, 1.5 m in length and 0.028 m in diameter) were positioned relative to the participant. The reaching movement was initiated in response to a verbal cue and recorded once using a three-dimensional optical motion analysis system (MA-5000; Anima Co., Japan, Tokyo, Japan) at a sampling frequency of 250 Hz. Simultaneously, the movement was recorded using two video cameras positioned anterolateral and posterolateral to the measured upper limb. The kinematic data obtained from this trial served as the reference kinematic data for the main experiment. The recordings from the two camera views were subsequently edited into an observation video of the reference movement, which was presented to participants in the main experiment. A humanoid model was then constructed based on the anthropometric measurements of the participant from whom the reference data were obtained (Fig. 2).

Fig. 2.

Schematic illustration of the reaching task and humanoid model.

Participants reproduced a reaching movement in which the fingertip marker moved from a start position to a circular target while avoiding a vertical obstacle bar. The humanoid model was constructed based on anthropometric measurements. The shoulder joint had three degrees of freedom (DOF), whereas the elbow and radio-ulnar joints had one DOF each. The wrist and finger joints were fixed. The green dotted line indicates the fingertip trajectory, and the red circles indicate the start and contact positions.

The main experiment included 36 individuals: 12 skilled physical therapists, 13 novice physical therapists, and 11 individuals with no experience in movement instruction. Participants were eligible if they had no neurological or orthopedic disorders affecting movement and were 21 to 40 years of age. Skilled physical therapists had more than five years of clinical experience, whereas novice physical therapists had less than one year of experience. Participants observed a video of the reference motion and were instructed to reproduce the movement of the humanoid model as accurately as possible in both the spatial and temporal aspects. Each participant completed three practice blocks, followed by a 15-minute rest, and then performed five test trials; the mean of the five trials was used for the analysis.

The humanoid model was constructed using aluminum frames and joints (G-fun, SUS Co., Ltd., Japan, Shizuoka, Japan) with a simplified joint structure that provided three degrees of freedom at the shoulder and one degree of freedom at the elbow and forearm, thereby approximating the upper-limb movement while maintaining structural stability.

Six kinematic parameters were calculated from the three-dimensional fingertip motion data of the humanoid model: movement time, trajectory length, peak velocity and its timing, positive and negative peak acceleration and their timing, and normalized jerk cost. Movement onset and termination were determined based on deviations from the stationary fingertip position, and the movement time was defined as the interval between these points. The trajectory length was calculated as the cumulative distance between successive coordinates, and velocity and acceleration were derived by differentiating the position data. The timing of each peak was normalized against the duration of the movement . The normalized jerk cost was calculated as an index of movement smoothness after normalization of the movement duration and displacement. For the waveform analysis, the Euclidean norm of the three-dimensional position, velocity, and acceleration signals was calculated and time-normalized to 101 points representing 0–100% of the movement duration.

Group differences were analyzed for each parameter. Normality was assessed using the Shapiro–Wilk test, and homogeneity of variance was assessed using Levene’s test. When these assumptions were satisfied, a one-way analysis of variance with Tukey’s post-hoc test was applied; otherwise, the Kruskal–Wallis test with Dunn’s post hoc test was used. Differences in continuous waveforms were analyzed using one-dimensional statistical parametric mapping with permutation-based inference. Statistical analyses were performed using R (version 4.5.0; R Foundation for Statistical Computing) and MATLAB (R2025a; MathWorks, Natick, MA, USA). The significance level was set at α=0.05.

RESULTS

In total, 36 individuals were enrolled in this study. During data processing, six participants were excluded because three-dimensional coordinate data were missing (two in the skilled physical therapist group, three in the novice physical therapist group, and one in the control group), leaving 10 participants per group in the final analysis. No significant differences in age, height, or body weight were observed among the three groups (Table 1).

Table 1. Participant characteristics.

Variable	sPT (n=10)	nPT (n=10)	Control (n=10)
Age (years)	31.5 ± 3.2	23.1 ± 0.9	31.2 ± 2.8
Height (cm)	169.1 ± 10.3	169.1 ± 8.4	165.0 ± 10.8
Weight (kg)	61.4 ± 7.7	63.2 ± 9.1	62.1 ± 11.9
Sex (male/female)	7/3	6/4	5/5
Clinical experience (years)	8.7 ± 1.3	1.0 ± 0.0	−

sPT: skilled physical therapists; nPT: novice physical therapists.

The fingertip trajectory length error met the assumptions of normality and homogeneity of variance, and no significant group differences were observed (F(2, 27)=1.30, p=0.29). Movement time error showed a significant group difference (χ²(2)=15.04, p<0.001). Post-hoc comparisons indicated that the skilled physical therapist group showed smaller errors than the novice physical therapist and control groups (both p<0.05). For peak velocity error, no significant group differences were observed (χ²(2)=4.19, p=0.123). In contrast, peak velocity timing error showed a significant group difference (χ²(2)=6.60, p<0.05). Post-hoc comparisons indicated that the skilled physical therapist group had fewer errors than the control group did (p<0.05). No significant group differences were observed in positive peak acceleration error (χ²(2)=4.11, p=0.128) or its timing error (χ²(2)=3.26, p=0.197), nor in negative peak acceleration error (χ²(2)=2.79, p=0.247). However, negative peak acceleration timing error showed a significant group difference (χ²(2)=7.00, p=0.030), and post-hoc comparisons indicated smaller errors in the skilled physical therapist group than in the control group (p<0.05). No significant group differences were observed in normalized jerk cost error (χ²(2)=5.23, p=0.073).

Continuous waveform analysis using statistical parametric mapping revealed no significant group differences in the fingertip trajectory, velocity, or acceleration profiles across the normalized movement cycle (Table 2, Fig. 3).

Table 2. Kinematic performance errors and group comparisons among sPT, nPT, and control participants.

Variable	sPT (n=10)		nPT (n=10)		Control (n=10)
Fingertip trajectory length error (m)	0.17±0.15		0.15±0.05		0.22±0.09
Movement time error (s)	0.24	[0.17–0.44]	0.57	[0.34–1.08]*	1.03	[0.78–1.81]***
Peak velocity error (m/s)	0.18	[0.10–0.23]	0.32	[0.15–0.50]	0.32	[0.24–0.52]
Peak velocity timing error (%)	4.20	[3.61–6.14]	5.83	[4.11–7.77]	7.04	[5.13–10.70]*
Positive peak acceleration error (m/s2)	0.51	[0.50–1.20]	1.07	[0.79–1.57]	1.33	[0.67–1.68]
Positive peak acceleration timing error (%)	8.44	[6.06–10.00]	8.78	[6.21–9.58]	8.09	[7.32–10.46]
Negative peak acceleration error (m/s2)	0.36	[0.32–0.94]	1.02	[0.84–1.33]	1.24	[1.01–1.61]
Negative peak acceleration timing error (%)	6.87	[5.23–8.90]	10.79	[7.97–14.13]	12.04	[10.13–14.45]*
Normalized jerk cost error (a.u.)	13.21	[9.36–23.34]	39.21	[9.33–63.97]	66.85	[36.20–111.67]

Values are presented as mean ± SD or median [IQR]. Data are shown as mean ± SD for normally distributed variables and median [IQR] for non-normally distributed variables. *p<0.05, ***p<0.001 vs. the sPT group. sPT: skilled physical therapists; nPT: novice physical therapists; Control: individuals without experience in movement instruction.

Fig. 3.

Group differences in kinematic performance errors.

Boxplots with individual data points show movement time error (a), peak velocity timing error (b), negative peak acceleration timing error (c), and normalized jerk cost error (d) for the skilled physical therapist (sPT), novice physical therapist (nPT), and control groups. The boxes indicate the interquartile range, the central line indicates the median, and the whiskers represent the data range excluding outliers. Statistical comparisons are reported in Table 2.

DISCUSSION

This study quantitatively evaluated manual guidance skills based on motion observation and identified the kinematic characteristics associated with the expertise of physical therapists. Errors in movement time, peak velocity timing, and negative peak acceleration timing were smaller in the skilled physical therapist group than in the other groups, whereas no clear group differences were observed in the spatial trajectory measures or continuous waveform structures. These findings suggest that expertise in manual guidance may be more strongly associated with the temporal organization of movement than with spatial accuracy.

The smaller movement time error in the skilled group indicates greater temporal accuracy in reproducing the observed movements. This finding is consistent with motor control theories suggesting that relative timing is a key component of motor planning and can be refined through practice⁷⁾. Similarly, the differences in peak velocity timing and negative peak acceleration timing suggest that skilled therapists may regulate the temporal structure of movement more consistently. In particular, a smaller error in negative peak acceleration timing may reflect a more precise control of movement termination, which has been associated with the anticipatory regulation of movement dynamics⁸⁾.

In contrast, no significant group differences were observed in the trajectory length or overall waveform profiles of the position, velocity, and acceleration. These results suggest that the spatial characteristics of movement were largely shared across participants. This finding is consistent with those of previous studies, indicating that spatial trajectory patterns tend to be relatively invariant, whereas expertise-related differences are more evident in the temporal aspects of movement^{9, 10)}. The constrained reaching task used in this study involved limited the degrees of freedom, which may have reduced variability in spatial performance.

Although normalized jerk cost error did not show a statistically significant group difference, the skilled physical therapist group tended to exhibit smaller errors and less variability than the other groups did. This tendency may reflect more consistent smoothness of movement during manual guidance.

The experimental task used in this study was intentionally designed to isolate one fundamental component of manual guidance skills, namely the ability to accurately reproduce an observed movement while minimizing therapist–patient interaction. In clinical practice, physical therapists perceive essential movement characteristics and provide appropriate manual guidance according to therapeutic goals. For example, therapists often guide the timing of weight transfer during sit-to-stand training, coordinate limb movements during gait training, and facilitate upper-limb reaching in patients with neurological disorders. In these situations, accurate temporal control may be particularly important for providing effective manual guidance. The present findings suggest that temporal aspects of movement may represent one fundamental component of manual guidance expertise.

This study has some limitations. First, the humanoid model did not fully replicate the complex interactions observed in clinical settings. Second, although no significant differences in height or body weight were observed among the three groups, this study did not evaluate the influence of anthropometric differences between therapists and the humanoid model, including upper-limb length and body proportions. Third, the task was limited to a single reaching movement with constrained degrees of freedom, which may limit generalizability. Fourth, the sample size was relatively small. Future studies using more complex tasks, larger samples, and humanoid models with varying anthropometric characteristics are needed to confirm these findings.

In conclusion, the results suggest that expertise in manual guidance is characterized primarily by more accurate control of the temporal structure of movement rather than by differences in spatial trajectory reproduction. These findings provide preliminary kinematic evidence of the characteristics of skilled manual guidance.

Conference presentation

Parts of this study were previously presented at the World Confederation for Physical Therapy Congress 2019 (Geneva, Switzerland), the 24th Annual Meeting of the Japanese Society of Physical Therapy Fundamentals (Niigata, Japan), and the 25th Annual Meeting of the Japanese Society of Physical Therapy Fundamentals (held online, hosted in Sendai, Japan).

Funding

This work was supported by JSPS KAKENHI Grant Number JP18K17729 and JP25K06524.

Conflict of interest

The authors declare no conflicts of interest.

Funding Statement

This work was supported by JSPS KAKENHI Grant Number JP18K17729 and JP25K06524.