In "Blade Runner 2049", the delicate scene where the replicants use mechanical hands to play the piano was once a human's imagination of future technology. Today, the dexterous hand is turning this imagination into reality - it is not a simple mechanical gripper, but a "ultimate robotic actuator" integrating three core systems of drive, transmission and perception. It is a crucial bridge connecting the digital world and the physical world.
The human hand, as the most sophisticated "tool" in nature, has 23 degrees of freedom and can perform a wide range of operations from grasping heavy objects to picking up embroidery needles. Its movement function accounts for 54% of the entire body. The design of the dexterous hand is a biomimetic replication of the human hand, and it must meet two core conditions: it should allow objects to adjust their positions freely in three-dimensional space and also limit the object's degrees of freedom through multiple contact points to prevent it from slipping. High-end dexterous hands have 15 to 42 degrees of freedom, gradually approaching the operational limit of the human hand, enabling robots to upgrade from "repeating pre-set actions" to "adapting to handle complex tasks".
Compared with traditional end-effectors, the dexterous hand has achieved a comprehensive leap: the repetitive positioning accuracy can reach 0.05-0.17 millimeters (equivalent to 1/7 of the diameter of a hair strand), capable of completing micrometer-level assembly; the grasping force can be adjusted from 1 gram (the weight of a business card) to 30 kilograms, capable of handling both eggs and heavy objects; it can be adapted to various objects such as spheres and irregular parts without changing the structure, and can operate stably in complex environments such as darkness and narrow spaces. Data from a certain new energy vehicle manufacturer shows that the assembly efficiency of the production line equipped with the dexterous hand has increased by 40%, and the failure rate has decreased by 65%.
Core of the work: The human-like closed loop of "perception - decision - execution"
1. Perception Layer: Equipped with "eyes" and "skin", to understand the physical world
The perception layer serves as the information gateway for the dexterous hand, similar to the visual and tactile systems of humans. Visually, the high-precision 3D camera can capture 3D point clouds with an accuracy of 0.02 millimeters, and the fine patterns on the screw surface can be clearly discerned; tactually, the flexible tactile sensors are attached to the fingers and palm, capable of distinguishing 20 types of surface textures and perceiving minute force changes of 0.01 Newton (equivalent to the weight of a single hair).
Take the capture of a cake as an example: The visual sensor first identifies the size, shape and cream distribution of the cake, while the tactile sensor anticipates the softness in advance. It can even determine the freshness of the cake through force feedback. The dexterous intelligent DexHand 021 dexterous hand is equipped with 23 multi-type sensors and is deeply integrated with the visual system, providing comprehensive data support for subsequent decisions.
2. Decision-making Level: Equipped with an "intelligent brain" to achieve precise judgment
The decision-making layer is the core of the dexterous hand, similar to the cerebral cortex of the human brain, responsible for "thinking things through before acting". In the early days, the dexterous hand relied on preset programs, and it might fail when handling a different object; nowadays, it is equipped with multimodal large models that integrate vision, touch, and physical rules to generate the optimal strategy.
The scene of picking up eggs best demonstrates its intelligence: The decision-making layer first identifies the properties of the object through vision, calls on the built-in database to know that the egg's compressive strength is less than 5 Newtons, and then plans the "three-finger 120° uniform force application" scheme. It also reserves 0.1 Newton of anti-slip redundant force to predict the risk of sliding. In a 3C electronics factory, it can identify 0.3-millimeter-thick wiring, through tactile perception of the force limit (beyond 0.1 Newton will cause breakage), and deduces the operation logic of "horizontal insertion and fine adjustment alignment". Even if the vision is blocked by 70%, it can still complete the task by relying on tactile sensation.
3. Execution Layer: Comparable to "the human cerebellum", achieving millimeter-level accuracy
The execution layer converts decisions into precise actions. The motion controller adjusts joint parameters at a frequency of 1000 hertz (1000 times per second), with a response speed that is five times faster than that of humans. The Bytai Intelligent D22 Pro dexterous hand uses a dedicated protocol to compress the communication delay between decisions and the control system to 2 milliseconds, with 500 data exchanges per second, comparable to the speed of human nerves.
If the contact point during egg grasping deviates by 0.1 millimeter, the execution layer will correct the angle within 0.001 seconds; once the tactile sensor detects a tiny crack on the eggshell (indicating abnormal force feedback), it will immediately switch to the "two-handed holding mode". This dynamic compensation capability enables the dexterous hand to break free from the label of "mechanically rigid".
The Source of Power: Each of the Four Driving Technology Schools Demonstrates Its Own Special Skills
The driving method directly determines the precision, weight and applicable scenarios of the dexterous hand. Currently, there are four main technological routes:
? Motor drive:It can be regarded as "industrial muscle", achieving 0.1 millimeter-level displacement control through micro motors and harmonic reducers, with a large output torque. The hand of the Boston Dynamics Atlas robot uses this technology. Each finger has 3 degrees of freedom, with a grasping force of 50 Newtons. It is suitable for high-load scenarios such as industrial assembly, but the transmission components often make the weight exceed 1 kilogram.
? Hydraulic/Pneumatic Drive:As "fluid muscles", they utilize fluid pressure to drive the movement of elastic chambers and are inherently compliant. Festo pneumatic fingers weigh only 80 grams and can safely grasp eggs. The MIT soft robotic hand can operate in a temperature range of -20°C to 80°C and is suitable for applications such as food packaging and medical minimally invasive procedures. However, it requires an external pump station, which affects its integration.
? Shape memory alloy drive:The "temperature-controlled muscle" moves by leveraging the temperature memory effect of nickel-titanium alloy. Without any transmission components, its weight is reduced to less than 200 grams. The humanoid finger prototype from Tsinghua University embeds the alloy wire in silicone, with the minimum bending radius being 5 millimeters. It is suitable for use in prosthetics and other wearable devices, and only the issue of energy efficiency needs to be addressed.
? New type of drive:Electrostatic drive enables the realization of 0.5-millimeter micro-joints that can grasp 10-micron cells; light-responsive polymer fingers can bend within 0.1 seconds under laser irradiation, opening up possibilities for micro-surgical robots.
From the Laboratory to the Industrial Sector: The Groundbreaking Implementation of Dextrous Hands
After half a century of evolution, the dexterous hand has moved from the laboratory to practical applications in multiple fields, reshaping the efficiency boundaries of various industries.
Industrial Manufacturing:During the entire year, the Shengshi robot has delivered nearly 2,000 sets of five-finger dexterous hands. These hands have been used in the production lines of new energy vehicles to perform precise operations such as wire arrangement and chip installation, replacing 3 skilled workers and achieving a nearly zero error rate.
Medical rehabilitation:The 19-degree-of-freedom dexterous hand developed by the University of Science and Technology of China weighs only 0.37 kilograms. It is driven by shape memory alloys and helps amputees perform actions such as combing their hair and writing. The tactile feedback enables patients to regain the "grasping sensation".
Extreme Operations:The Shadow Dexterous Hand, equipped with 24 degrees of freedom and force sensors, can perform tool operations in nuclear waste disposal sites, thereby preventing humans from being exposed to dangerous environments.
Household Services:The dexterous hand equipped with an end-to-end decision-making architecture can autonomously identify irregular-shaped water cups and generate a "palms supporting the bottom + two-finger assistance" grasping strategy, providing the core support for the elderly care companion robot.
The Future Has Arrived:The next step in the evolution of dexterous hands
Currently, the dexterous hand is making rapid progress towards "lightweighting, integration, and intelligence": its weight is moving towards being under 500 grams, and the "motor + shape memory alloy" hybrid drive will balance strength and flexibility; the continuous training of multi-modal large models enables it to understand more complex scenarios and even predict human operation intentions.
From Watt's mechanical claws to the current intelligent dexterous hands, humans are using technology to replicate their own dexterity while also redefining the way machines interact with the world. When the "hands" of robots become sufficiently flexible, future industrial production, medical services, and family life will all witness more intelligent and warmer changes.
