Humanoid robots touch the world, starting from 2000 sensors | Product Observation
Author｜Huang Nan
Editor｜Yuan Silai
As the most flexible part of the human body joints, among the 206 human bones, the hand joints account for one-fourth. This also means that today, as humanoid robots are increasingly approaching human intelligence, in order to enable them to land in specific scenarios and complete tasks, the perception and control of a pair of dexterous hands play an important role in the breakthrough of the robot's capabilities.
Currently, the dexterous hand operation is mainly based on the visual solution. Based on visual sensors such as high-definition cameras and cameras, the image data of the surrounding environment is captured and processed to construct a three-dimensional structure, which can provide real-time visual feedback to help the humanoid robot accurately identify, locate and operate.
However, in a complex environment, due to the influence of factors such as lighting, occlusion, and perspective distortion, it may cause incomplete perception and information distortion of the visual sensor. The tactile solution can effectively make up for this limitation.
By sensing the contact force and contact state between the robot and the object in real time, the information feedback received by the tactile sensor is more timely and accurate, which helps the dexterous hand adjust its operating state, such as the grasping force, posture, or the trajectory of the hand movement, to ensure the task is completed stably and accurately. When dealing with soft, fragile or irregularly shaped objects, the feedback of the tactile sensor is particularly important.
With the increasing perception requirements of the dexterous hand for the tactile sensor, the traditional tactile sensors with low accuracy, low sensitivity, poor resolution and high cost cannot meet the market demand and are not suitable for large-scale deployment; while the high-end tactile sensors are monopolized by European and American countries, occupying more than 80% of the market. As the underlying technology of humanoid robots and other advanced manufacturing industries, accelerating the domestic substitution of tactile sensors has become a key issue.
PaXini Tech., which Hardcore has been paying close attention to for a long time, is an innovative company of tactile sensors and humanoid robots with multi-dimensional tactile sensing technology as the core. Its core products cover the full-line layout of multi-dimensional tactile sensors, multi-fingered dexterous hands, and humanoid robots.
At the 2024WRC this year, PaXini Tech. officially released its second-generation multi-dimensional tactile humanoid robot TORA-ONE, the second-generation multi-dimensional tactile dexterous hand DexH13, and the multi-dimensional tactile sensor PX-6AX GEN2 based on the ITPU (Intelligent Tactile Processing Unit) technology.
It is a reality that needs to be acknowledged that, matching the current level of robot intelligence, the operational ability of the upper body covers a large part of its applications, such as assembly on the factory assembly line, robotic arm welding, and handling. When humanoid robots enter more landing scenarios, it will involve a large number of refined operation tasks in human life, and it needs to have more abundant perception, reliable decision-making and stable execution capabilities.
Relying on the tactile sensor, PaXini Tech. has the full-chain production capacity from core components, dexterous hands to the whole machine manufacturing of humanoid robots. The products are widely used in many fields such as industrial manufacturing, precision processing, medical health, and warehousing and logistics.
More Flexible and Freer
From the appearance, the second-generation multi-dimensional tactile humanoid robot TORA-ONE released by PaXini Tech. has upgraded the degrees of freedom of both arms and dexterous hands. The more degrees of freedom a humanoid robot has, the better its flexibility and versatility.
The TORA-ONE consists of a total of 47 degrees of freedom, including 21 for the body and 26 for the four-fingered bionic dexterous hand, with a single-arm load capacity of 5 kilograms. Among them, the four-fingered bionic dexterous hand DexH13 and the multi-dimensional tactile sensor PX-6AX GEN2 it is equipped with are the two core products launched by PaXini Tech. this time.
In the previous generation of robots represented by industrial robots, the end effector, according to the needs of the application scenario, mostly exists in the form of a gripper in its hand, and can only perform a single type of task on the production line, with low versatility.
As a multi-dimensional tactile dexterous hand, DexH13 is also the first four-fingered bionic dexterous hand on the market that integrates the bimodal of tactile and visual. It has 13 active degrees of freedom in a single hand; it can not only complete industrial tasks such as lifting, clamping, and welding, but also simulate humans to perform complex hand movements such as grasping and rotating, and performs well in flexible tasks.
Multi-modal perception, flexible operation
In order to increase the tactile perception sensitivity and hand grasping force feedback, PaXini Tech. has equipped nearly 2000 self-developed and produced high-precision tactile sensors PX-6AX GEN2 on the hands of DexH13, which can perform 15 kinds of multi-dimensional tactile perceptions such as pressure sensing, friction, and texture at ultra-high resolution, achieving a full-range 0.01N high-precision measurement; and provides a fine temperature compensation instrument, in addition to three-dimensional and six-dimensional forces, it can also perform tactile perception of multi-dimensional physical information such as material and temperature.
In terms of vision, the hands of DexH13 are integrated with an 8-megapixel high-definition AI hand-eye camera. Based on the advanced zero-shot pose estimation visual algorithm, combined with the 5 cameras and 2 depth cameras of TORA-ONE, it can identify and grasp the six-dimensional pose of the object and adapt to a variety of complex environments and tasks.
Estimate the object pose and grasping posture based on the VTLA multi-modal perception model, and evaluate the precise grasping point
In the lower limbs of the TORA-ONE, its chassis and torso adopt a highly modular structural design with strong stability. The folding design of the waist can be flexibly extended, and the height range is from 1.46 meters to 1.86 meters to meet the operational requirements of various three-dimensional space heights in industrial manufacturing, medical and health care, warehousing and logistics, etc.
It is equipped with a highly mobile universal mobile chassis with a 360-degree omnidirectional movement capability. Even in a complex workshop layout or dynamic working environment, it can complete adaptive rapid movement with a maximum speed of 1m/s and a battery life of up to 8 hours, providing a reliable and stable power support for high-intensity continuous operations in industrial production.
At the same time, the mobile chassis of TORA-ONE is also equipped with a LiDAR SLAM positioning and navigation system, which can accurately position in a complex three-dimensional environment, accurately identify various objects and location information, quickly plan the optimal action path, and accurately avoid obstacles, greatly improving production efficiency and operational stability.
Efficient operation across scenarios
Today, although traditional large-scale industrial automation equipment has improved production efficiency to a certain extent, due to the lack of cognitive ability in complex scenarios, the application space is limited, and it is mostly used to solve specific simple tasks, and it is difficult to respond quickly and flexibly to delicate and complex manual products and complex and changeable production needs.
The products with increasingly complex and highly customized needs have increased significantly, and a flexible and accurate response to batch customized orders is the key.
Taking the industrial scene as an example, in the fine assembly and delivery link, TORA-ONE based on the tactile sensor and the visual-tactile multi-modal perception model, can quickly sense and determine the shape, size and material of the parts, and then use the dexterous hand to complete the assembly of the precision parts and transport them to the designated location.
It adopts advanced kinematics modeling and control algorithms, which can complete stable and high-precision operation capabilities under high degrees of freedom movement, and safely and efficiently interact with surrounding equipment and the environment, flexibly adapting to various operations such as flexible grasping, assembly, loading and unloading of various workpieces.
With the improvement of the intelligent level and motor function of TORA-ONE, it finally realizes the transition from a single basic labor to more complex flexible tasks.
Deep learning, continuous optimization
Early Self-Blood-Making Ability
According to the latest report data from Omdia, it is expected that by 2027, the global shipment of humanoid robots will exceed 10,000 units, and it will reach 38,000 units by 2030, with a compound annual growth rate of 83% from 2024 to 2030.
However, as a large intelligent mobile terminal integrating AI algorithms, motion control, hardware structure, supply chain, etc., the humanoid robot is currently recognized as the most difficult product, with nearly 5,000 parts alone, and the system engineering is extremely large.
The humanoid robot shows a huge market demand. But at this stage, the industry still has technical, cost and other issues to be solved, and some enterprises are still in the stage of product research and development and testing, and the commercialization process is still in the early stage.
Therefore, many enterprises either choose to take the more mature commercialization tracks such as educational research, service robots, or core components as the entry point. This part of the business can bring a certain stable income to the enterprise in the early stage and maintain the blood-making ability.
In terms of hardware, in addition to the reducer, servo system and controller with the highest technical difficulty, the controller, sensor, battery system, etc. used in the humanoid robot have a large degree of repetition with the applications in the fields of industrial robots and new energy vehicles.
Take the tactile sensor independently developed by PaXini Tech. as an example. In terms of products, for the consumer-level robot and smart home field, PaXini Tech. has launched the PX-3AX tactile sensor, which focuses on high cost performance and can achieve multi-scenario, high-precision and multi-dimensional tactile perception.
At present, multi-dimensional tactile sensors have been widely used in many refined operation scenarios. Based on massive multi-dimensional tactile data, the reliability of equipment operation can be effectively improved. And the algorithm driven by the multi-modal model and the AI's autonomous learning ability can enhance the universality of the sensor in complex task programs and multiple scenarios.
Smart force control, high-performance arm-hand combination
At present, PaXini Tech.'s multi-dimensional tactile sensors have been applied in industrial robots, surgical robots, mobile robots, service robots and intelligent cockpits.
The official online mall has also been officially launched today. Among them, PX-6AX GEN1 has been on sale, and another professional-level robot ITPU PX-6AX GEN 2 is also planned to be launched and sold one week later.
Next, PaXini Tech. will continue to promote the application of multi-dimensional tactile sensors in more scenarios, explore potential markets, and accelerate the landing process of dexterous hands, further exploring the possibility of industrialization from upstream components, core equipment to the whole machine of humanoid robots.