A joint research team comprising Peking University, Beijing Institute for General Artificial Intelligence (BIGAI), and other institutions has developed the world’s first robotic hand system featuring both full-hand high-resolution tactile perception and complete dexterous motion — the “Full-Tactile Anthropomorphic Cybernetic Hand” (F-TAC Hand). The results were published on June 9 (Beijing time) in Nature Machine Intelligence, a leading international journal. What technical challenges did this breakthrough overcome? And how will it advance the development of embodied intelligence? The Global Times interviewed members of the research team for insights.

Mimicking Human Touch: How the F-TAC Hand Works

The human hand consists of 27 bones and 34 muscles, providing 24 degrees of freedom—meaning it can move in 24 independent ways across its various joints. This high level of mechanical complexity enables both strength and delicacy, making the hand an ideal model for advanced robotics and embodied intelligence research. Understanding the human hand’s capabilities remains a frontier challenge in both robotics and embodied AI research. Grasping objects involves two core faculties: tactile feedback and motor control. Historically, integrating these two has been one of the most formidable technical obstacles in robotic systems.

“The human hand’s tactile system relies on two key components: a dense array of tactile receptors embedded in the skin and a neural processing mechanism in the brain to interpret these signals. The F-TAC Hand mimics this architecture,” said Zhao Zihang, the first author of the paper and a Ph.D. candidate at the Institute for Artificial Intelligence, Peking University.

The F-TAC Hand integrates 17 high-resolution tactile sensors in six distinct configurations. Uniquely, the sensors serve both as sensory elements and as structural components. This enables an unprecedented tactile coverage area without compromising flexibility.

The sensors cover 70% of the palm surface and achieve a spatial resolution of 0.1 mm, providing approximately 10,000 tactile pixels per square centimeter—far surpassing the capabilities of current commercial robotic hands. For example, the advanced Shadow Hand offers only single-point force sensing at the fingertips. This innovation enables real-time contact perception and rapid grip adjustments, significantly enhancing the robotic hand’s stability when operating in uncertain environments.

Overcoming Bottlenecks in Human-like Dexterity

F-TAC Hand solves the longstanding challenge of integrating tactile feedback with dexterous motor control through its unified sensor-structure design.
“From a hardware perspective, adding tactile sensors without sacrificing joint flexibility is a major obstacle. Moreover, existing tactile sensing technologies still fall short in terms of coverage, resolution, and durability for practical use,” explained co–first author Li Yuyang, a Ph.D. candidate at BIGAI.

Even with high-resolution tactile perception, controlling a high-degree-of-freedom robotic hand to perform complex human-like tasks remains a major challenge due to the sheer volume of data and coordination required. To address this, the team developed a probabilistic algorithm that generates diverse human-like grasp strategies. “It replicates 19 commonly observed human grasp types,” Li noted.

He further elaborated on the adaptive intelligence of F-TAC Hand: “Simultaneous multi-object grasping is a crucial benchmark for evaluating dexterity. Unlike simple two-finger pinch grasps, grasping multiple objects requires full-hand contact detection and precise motion adjustment to ensure stability.”
Advancing Embodied Intelligence

Advancing Embodied Intelligence

“This research is not just a technical milestone—it offers a new perspective on the nature of intelligence,” said corresponding author Zhu Yixin, Assistant Professor at the Institute for Artificial Intelligence, Peking University.

“Human intelligence is deeply rooted in the body’s perceptual abilities. Tactile experience, especially through the hands, is fundamental to how we understand and interact with the world. The F-TAC Hand demonstrates that rich sensory input is essential for advancing machine intelligence.”

Zhu added, “We will continue to integrate tactile perception with robotic control and explore more intelligent forms of embodied interaction, laying the groundwork for artificial general intelligence.”

He emphasized that the F-TAC Hand opens new research frontiers in embodied AI. “Combining high-fidelity physical perception with intelligent control systems is a crucial path toward more advanced machine intelligence. This study not only pushes robotics forward but also deepens our understanding of intelligence itself. It offers valuable insights for the next generation of AI systems and promotes the integration of embodied AI across industries—powering China’s high-quality development and contributing to new productive forces.”

Editor: LQQ