Abstract
<jats:p>Musculoskeletal disorders from repetitive surface-finishing tasks remain a leading cause of occupational disability. Robotic assistive devices can reduce operator muscular load, yet existing control strategies typically assume known surface geometry, decouple force regulation from human guidance, or treat all contact directions identically. This systematic review investigates how a control strategy for a mechanically coupled assistive device can maintain desired contact force along the surface normal while providing power-assist in tangential directions, when surface geometry is unknown and must be estimated online from force-motion interaction. The research question is decomposed into five components: online surface estimation (C1), physical human-robot interaction (C2), haptic shared control (C3), anisotropic force control (C4), and ergonomic load reduction (C5). Following PRISMA 2020 guidelines, Scopus and IEEE Xplore were searched on 8 January 2026 using seven structured queries, supplemented by citation chaining from 23 seed papers. Each paper was scored on a five-component rubric (ABSENT / IMPLICIT / EXPLICIT) and assigned to quality tiers (1 DIAMOND, 27 GOLD, 31 SILVER, 5 RIVAL, 7 FOUNDATIONAL). From approximately 12,600 identified records, title-and-abstract screening retained 72 reports for full-text assessment, of which 71 studies were included. Component coverage is uneven: C2 is addressed by 49 papers (69%), C3 by 44 (62%), C4 by 38 (54%), C5 by 30 (42%), and C1 by 25 (35%). A progressive funnel analysis reveals that only 7 papers combine C1+C2+C4 at IMPLICIT level or above, and 0 of 71 combine all five components with genuine online surface estimation from force-motion interaction. The gap is structural: surface estimation methods assume autonomous operation, pHRI methods assume known geometry, and ergonomic validations use simplified surfaces. A search update run on 12 July 2026, immediately prior to posting, screened 113 newly indexed records against the same rubric, with every sought report retrieved and assessed in full text; none combines these components, and the gap stands. The review confirms a specific integration gap: the five required capabilities have each been demonstrated individually, but the assumptions underlying each are mutually incompatible. Limitations include single-reviewer screening and coverage restricted to two databases, partially mitigated by citation chaining. The protocol was not prospectively registered. No external funding was received.</jats:p>