New survey maps robotic dexterous hand intelligence across hardware, learning, and evaluation

A new survey paper titled 'Towards Robotic Dexterous Hand Intelligence: A Survey' offers a holistic review of the rapidly evolving field of robotic dexterous hands, which are central to contact-rich manipulation. Authored by Weiguang Zhao and five other researchers from multiple institutions, the survey systematically organizes research across four complementary aspects: hardware-level analysis, control and learning methodologies, data and evaluation practices, and future challenges. The hardware section delves into actuation, transmission, perception, and representative hand designs, emphasizing trade-offs between force capability, compliance, bandwidth, integration, and system complexity.

The survey also categorizes control and learning methods by major paradigms (e.g., model-based, learning-based, hybrid) and traces their chronological evolution, enabling readers to see how techniques have advanced. It consolidates datasets, modality designs, and evaluation practices, connecting methodological progress with how systems are trained and benchmarked. By linking hardware, algorithms, data, and assessment, the paper provides a structured understanding of the field and identifies the most important open challenges—such as generalization, sim-to-real transfer, and achieving human-level dexterity. This survey serves as a critical reference for researchers and engineers aiming to advance robotic hand intelligence.