A systematic review of observational practice for adaptation of reaching movements

Motor learning encompasses acquiring new skills and adapting existing ones. While physical practice is typical, observational practice (observing a model performing the task) also offers benefits. Reaching movements are crucial for daily activities and rehabilitation, and are also relevant to robotics. Previous research on observational learning has mostly focused on complex skills, while a systematic review of simpler reaching tasks relevant to daily life was lacking. This study aimed to determine whether observational practice of reaching movements leads to similar learning rates and neural mechanisms as physical practice, and to identify factors determining learning success. The potential of observational learning as a substitute or complement to physical practice is particularly relevant when novel complex movements are being learned, movement execution is restricted, or unskilled practice carries severe consequences. The effectiveness of observational learning depends on factors such as task complexity, skill level, attention, and motivation of the learner.

Research on observational learning dates back to Bandura's social learning theory, which posits that observed behavior is encoded symbolically and stored as a cognitive memory representation, involving attention, retention, reproduction, and motivation. The discovery of mirror neurons provided a physiological explanation. While direct evidence of mirror neurons in humans is lacking, neuroimaging studies reveal mirror-like behavior in brain networks, including the inferior frontal gyrus, premotor cortex, sensory cortices, and cerebellum. This suggests that observational and physical practice might modify and consolidate task-specific sensorimotor networks similarly. Tasks used to investigate observational motor learning differ in skill knowledge (new vs. known skills), learner expertise, and extremity involved. Ramsey et al. distinguish sequence learning (sequential tasks) and motor adaptation (adjusting to new constraints). Motor adaptation involves updating inverse (motor commands) and forward (sensory consequences) models. Aftereffects (errors after adapting to a force-field and returning to the null-field) are a hallmark of successful adaptation.

A systematic review was conducted using PsycInfo, PubMed, Scopus, and Web of Science databases. The search terms combined "motor" or "movement," "learn" or "adapt," "observ" or "imitat," and "reach" or "aiming." After removing duplicates, 10,620 studies remained. Title and abstract screening yielded 32 full-text articles for eligibility assessment based on PICO and PRISMA guidelines. Studies were included if participants were healthy adults (18+ years), used a reaching or aiming task with at least one pure observational practice condition, were published in English peer-reviewed journals, and focused on reaching tasks. Studies on sequence learning, those lacking observational conditions, and non-human studies were excluded. Data extraction included study design, task type, participant demographics, reaching task, experimental conditions, stimulus presentation, measurement device, and outcome variables. Two independent reviewers extracted data, with discrepancies resolved by discussion. The revised Cochrane risk-of-bias tool (RoB 2.0) assessed study quality in five domains. Standardized mean differences (Cohen's d) were calculated and synthesized using R and the metafor package. The GRADE approach assessed the certainty of evidence.

The final analysis included 18 studies involving 1029 young adults (mostly right-handed). Reaching tasks used robotic manipulanda (force fields), graphic tablets (visuomotor rotations), or exoskeletons. Observational practice led to adaptation in both paradigms, with effect sizes ranging from d = -0.5 to -5.07 (congruent) and d = 0.55 to 3.36 (incongruent). However, physical practice showed larger effects (d = 4.38) in the three studies comparing both. Aftereffects were weaker or absent after observational practice, suggesting differences in learning mechanisms. Observational practice led to better explicit knowledge of the task constraints. The visualization of task-errors was crucial for learning gains, and observing high-error trials was more beneficial than observing low-error trials. Mixed schedules (observational and physical practice) showed varying effects, sometimes outperforming isolated practices. Neurophysiological studies indicated the involvement of M1, S1, V5/MT, PMd, SPL, and cerebellum in observational learning, similar to physical practice. Inhibition of M1 reduced learning effects. However, only five studies investigated neural correlates, limiting generalizability. Risk of bias was high or some concerns in most studies, and publication bias was suspected due to the dominance of two research groups in the included studies. The overall certainty of evidence was moderate for young adults and the specific tasks.

This review showed observational practice improves reaching performance, though less so than physical practice. The absence of aftereffects suggests primarily explicit rather than implicit learning, potentially affecting inverse but not forward models. However, this might depend on task characteristics (force-fields versus visuomotor rotations). The visualization of task errors and the involvement of somatosensory information processing seem to play a crucial role in observational learning. Mixed schedules may integrate explicit and implicit strategies, offering advantages. While neurophysiological studies point to involvement of similar brain regions as physical practice, more research is needed to explore this thoroughly.

Observational practice effectively facilitates motor adaptation in reaching tasks, but its mechanisms differ from physical practice. Visualizing task errors is crucial for successful observational learning, and combined practice schedules might optimize learning outcomes. Future research should investigate neural correlates of both practice types directly, use more diverse tasks and participant populations, and address the role of forward models in observational learning.

The review's findings are limited by the methodological heterogeneity of the included studies, the predominance of data from two research groups using specific paradigms, and the relatively small number of neurophysiological studies. The generalizability to other populations and tasks might be limited.