A review of adaptive intelligence in tactile sensing robotic hands for human centered dexterous control

Abstract The development of dexterous hands that can accommodate and reason about contact uncertainty—rather than blindly following pre-planned trajectories—remains one of the most consequential unsolved problems in robotics engineering. This article reviews 115 studies published between January 2020 and December 2025, synthesising advances across the hand–arm pipeline in four areas: embodiment, tactile and multimodal perception (slip detection, contact-state estimation, texture recognition, local geometry, and stiffness inference), dexterous control and learning, and human-centered deployment. Several patterns recur consistently across the reviewed corpus. Tendon-driven systems remain the dominant hand architecture, yet many evaluations underreport the limitations imposed by friction, hysteresis, and calibration drift, which together erode repeatability in ways that matter for sustained deployment. Soft and compliant end-effectors broaden the safety envelope in human-shared settings but generally reduce the precision available for fine in-hand manipulation. Across the reviewed studies, vision–tactile fusion was frequently associated with stronger manipulation robustness than vision-only baselines; this advantage holds across task types, though the magnitude varies with sensing quality and controller design. Within the reviewed corpus, pure reinforcement learning pipelines remained predominantly confined to simulation or simplified hardware, while model-based and hybrid controllers appeared more consistently on real hardware under deployment-grade latency and safety constraints. Four gaps recur across all reviewed areas: principled coupling of semantic reasoning with tactile execution, credible safety certification for close human–robot proximity, the absence of standardised benchmark reporting, and unresolved long-horizon reliability. A deployment-oriented evaluation framework and practical readiness criteria are synthesised to guide future research priorities.