Psychomotor Skill Benchmarking Eval

psychomotor-skill-benchmarking-eval

Performance Benchmarking of Psychomotor Skills Using Wearable Devices: An Application in Sport — Pandukabhaya et al. (2024) (arXiv:2411.16168, 2024)

What this evaluates

Evaluates the objective quantification and benchmarking of psychomotor execution quality in sports using wearable IMU data. It maps raw 3D motion trajectories into a normalized performance space and uses unsupervised clustering to identify optimal movement patterns and detect technical deviations.

Datasets

• Table Tennis Forehand Stroke (IMU) — total 18; splits: test (-1)

Metrics

• Euclidean distance to ideal performance origin (primary) — range: other
  • Computes the Euclidean distance between a cluster centroid in the 5D performance space and the origin of ideal performance O' ≡ (1, 1, 1, 1, 1). Lower distance indicates closer alignment with optimal psychomotor execution.
• Mean Euler angle variation — range: other
  • Calculates the average yaw, pitch, and roll angles across the wrist, elbow, and shoulder joints for each cluster to visualize and compare movement kinematics against the benchmark.

Input / output format

Input: Time-series 3D motion data (Euler angles) captured by wearable IMU sensors attached to the wrist, elbow, and shoulder during table tennis forehand strokes.

Output: Cluster assignments for each stroke, centroid coordinates in the 5D performance space, and mean Euler angle trajectories per cluster.

Scoring recipe

# 1. Normalize raw IMU trajectories into 5D performance space using RBF cost functions
perf_space = [rbf(cost_func, trajectory) for trajectory in trajectories]
# 2. Cluster performance space (e.g., k-means with K=4)
clusters = kmeans(perf_space, k=4)
# 3. Compute centroid for each cluster
centroids = [mean(cluster_points) for cluster_points in clusters]
# 4. Calculate Euclidean distance to ideal origin O'=(1,1,1,1,1)
distances = [euclidean_dist(c, [1,1,1,1,1]) for c in centroids]
# 5. Identify benchmark cluster as the one with minimum distance
benchmark_cluster_idx = argmin(distances)

Common pitfalls

• The benchmark cluster is identified unsupervised via proximity to an idealized origin, not through expert-labeled 'perfect' strokes.
• Cost functions and performance space dimensions are highly domain-specific to table tennis forehand strokes and may not transfer to other sports without redefinition.

Evidence (verbatim from paper)

After selecting the 4 primary clusters, the centroid of each cluster was calculated using (19). Subsequently, the distance from the centroid of each cluster to the origin of ideal performance O' ≡ (1, 1, 1, 1, 1) in the performance space was calculated using (20). The results are summarized in Table 5. With the results obtained in Table 5 for the mean Euclidean distances, cluster q = 0 was identified as the benchmark cluster with the lowest distance of d_0 = 0.6284 from O' considering (21).

Citation

@misc{pandukabhaya2024performance,
  title={Performance Benchmarking of Psychomotor Skills Using Wearable Devices: An Application in Sport},
  author={Pandukabhaya et al. (2024)},
  year={2024},
  note={arXiv:2411.16168}
}

• arXiv: 2411.16168