Title
Architectural Trade-offs in Humanoid End-Effectors: A Multi-Attribute Decision Framework for Linkage versus Tendon-Driven Actuation
Authors
Abstract
Purpose: Humanoid robotic end-effectors are the physical interface of human–robot collaboration, yet the choice between linkage-driven and tendon-driven actuation is usually made as an isolated hardware judgment. This paper develops a reproducible multi-attribute decision-making (MADM) framework that couples the mechanical attributes of the hand to the human–AI decision dynamics formalised in the HACDU framework (Chan & Pan, 2026), so that actuation selection is treated as a utility-coupled human–machine decision problem rather than a purely mechanical one.
Design/Methodology/Approach: Five attributes — payload-to-weight ratio, dynamic responsiveness, physical compliance and impact resilience, maintainability/reliability, and human-in-the-loop override safety — are evaluated with an AHP-weighted TOPSIS model (Hwang & Yoon, 1981; Saaty, 1977). Because the candidate set contains exactly two architectures, we prove that the two closeness coefficients sum to one, reducing the ranking to a single scalar comparison. All reported scores were recomputed from the published parameter matrix, and weight uncertainty is audited with a Monte Carlo study (Dirichlet perturbations, N = 5,000) across three task profiles: deterministic precision manufacturing, collaborative eldercare, and balanced logistics.
Findings: Recomputed closeness coefficients are 0.74 (linkage) versus 0.26 (tendon) for precision manufacturing, 0.19 versus 0.81 for collaborative eldercare, and 0.45 versus 0.55 under balanced weights. Monte Carlo audit confirms the first two rankings as robust (rank-1 probabilities 0.98 and 0.998, respectively) but shows the balanced case to be genuinely contested (tendon rank-1 probability 0.65). A rank-reversal threshold is identified: once the compliance weight exceeds approximately 0.16 (residual weight split equally), tendon actuation overtakes linkage actuation.
Research Limitations/Implications: The parameter matrix synthesizes representative values from the cited hand literature, and the override-safety index is a defined bench protocol rather than measured data; physical prototype validation and human-subject trials remain future work. With two alternatives, the TOPSIS closeness is a relative, not absolute, utility measure.
Practical Implications: Robotics architects and technology executives obtain an auditable selection template — explicit weight vectors, consistency-checked AHP judgments, recomputed scores, robustness probabilities, and a single switching threshold — applicable across industrial manufacturing, eldercare, and disaster-response deployments.
Originality/Value: The paper bridges mechanical hand design and formal decision science: it corrects and fully operationalises the AHP–TOPSIS pipeline (including its two-alternative identity and rank-reversal behaviour), embeds HACDU belief fusion and override governance as a scored attribute, and reports the first Monte Carlo robustness audit for this selection problem.
Keywords
Humanoid robotics; end-effector actuation; linkage mechanism; tendon-driven hand; multi-attribute decision-making (MADM); AHP; TOPSIS; HACDU framework; human–robot collaboration; Monte Carlo sensitivity analysis
Classification-JEL
C44, C61, D81, L63, O32
Pages
55-72
How to Cite
Chan, L. C. W. J., & Pan, S. H. (2026). Architectural Trade-offs in Humanoid End-Effectors: A Multi-Attribute Decision Framework for Linkage versus Tendon-Driven Actuation. Annals of Artificial Intelligence, 2026(1), 55-72.