Produced by Tiger Sniff Think Tank

Author|Ji Yujie

Undoubtedly, at the 2024 World Robot Conference not long ago, embodied intelligent humanoid robots were the biggest highlight, and 27 humanoid robots were unveiled at the conference. They can run, jump, somersault, poetry, pour tea, fold clothes, and are proficient in all eighteen kinds of martial arts.

From simply showing off skills to working in the factory

For embodied intelligent humanoid robots, it is necessary to pay attention to the dexterous operation ability of the upper body and the stable movement ability of the lower body.

Let's take a look at the upper body operation ability first.

Songyan Power's robot HObbs simulated face can simulate human facial expressions in real time, making a variety of expressions such as opening the mouth, frowning, expressing anger, smiling, and disgust; The Zhiyuan Expedition A2-W robot demonstrated the ability of dynamic task orchestration, complex job execution, and two-arm collaborative operation. The Astribot S1 two-finger gripper can skillfully work on tea, fold clothes, and complete a series of difficult, long-sequence, and generalizable tasks.

Figure: Songyan Power Bionic Robot HObbs

Pictured: Zhiyuan Robot A2-W

Pictured: Stardust Intelligent AI Robot Assistant Astribot S1

SIASUN Centaur vacuum manipulator has two arms that are not affected by each other to independently pick and place wafers and complete wafer handling of different sizes and specifications; Pasini's second-generation multi-dimensional tactile humanoid robot TORA-ONE, with four fingers, can compare hearts, write calligraphy, play the piano, and play chess. The Alpha Bot 1S can not only make a fist, but also make coffee, learn to operate various daily electrical appliances, and realize flexible switching between multiple scenarios and tasks.

Pictured: SIASUN Centaur vacuum manipulator

Pictured: Pacini TORA-ONE comparing hearts

Pictured: Smart Square Alpha Bot 1S

The robot's lower body movement ability and stability have also improved greatly.

The CL-1 was tested at the exhibition site, and in the face of continuous kicking, it was still able to maintain the dynamic balance of the whole body, showing strong anti-interference ability and stability. The mechanical dog of Unitree Technology, with its four-legged dexterity climbing up and down, attracted everyone to marvel again and again.

Pictured: The CL-1 is constantly kicked

Pictured: Unitree Technology Robot Dog

Accelerated Evolution even held a football match on the spot, and its humanoid robot Booster T1 attracted attention with difficult movements such as throwing balls, kicking balls, and squatting with Ip Man, as well as push-ups that interacted with the audience. UBTECH simulated the automobile production line on site, and its humanoid robot Walker S series can autonomously move parts boxes from the warehouse floor to the conveyor belt, and record data in real time, realizing seamless connection with the warehouse system of the smart factory.

Pictured: Booster T1, a humanoid robot that has evolved faster

Picture: UBTECH Walker S series

Judging from the exhibition alone, humanoid robots are moving from show-off skills to service, and embodied intelligence seems to be about to usher in the iPhone moment. But in fact, in the real world, there are still many technical and product problems of humanoid robots that have yet to be solved.

The dexterous hand is the "configuration" C-position

At present, the development of general-purpose intelligent humanoid robots is gradually becoming a consensus in the industry. The biggest advantage of humanoid robots is that they have the ability to move in general. Among them, the lower limbs are responsible for stable movement, and the upper limbs are responsible for dexterous operation. Compared with the ability to "move" the lower limbs, to solve problems in specific scenarios, the "dexterous operation" ability of the upper limbs is more needed.

Since the rise of automatic control technology in the 50s of the 20th century, the "upper limb" robotic arm and industrial robots have a long history of development, but the ability is limited to mechanical, repetitive and relatively simple actions.

Therefore, the precise operation ability of the dexterous hand at the end of the upper limb has become the focus of attention in the current industry.

Dexterous hands can be roughly divided into electric grippers and multi-finger dexterous hands.

Compared with the electric gripper with low technical difficulty and high robustness, the development of multi-finger dexterous hands is gradually becoming a consensus. As a high-performance and high-flexibility end effector, the multi-finger dexterous hand can be used by customers in special and complex scenarios, and at the same time plays a more important role in improving the flexibility and ease of use of the robot.

As for the number of fingers, the jury is inconclusive. Most dexterous hands are five-fingered, but there are also four-fingered ones. In this regard, Xu Jincheng, CEO of Pasini, believes that "four or five fingers should be purpose-oriented, and four fingers may be enough in many cases, because human beings rarely use little fingers in real life." Pasini's

tactile dexterity is four fingers, with 13 degrees of active freedom, which can realize various complex human hand movements such as grasping, grasping, pinching, and pressing fingers to open and close.

Four or five fingers is not just about the quantity, but also about the technical difficulty and cost.

Ju Xiaozhu, head of the large model of the Beijing Embodied Intelligent Robot Innovation Center, said, "At present, dexterous hands are mostly simulated human hand joints, and a joint corresponds to a motor and a degree of freedom. The more fingers, the more joints, the more motors, and the higher the degrees of freedom. And within the finite size of a hand, the more motors and the more complex the structure, the greater the challenge to its reliability and sensitivity, and of course the higher the cost. "

It can be said that how to integrate multiple degrees of freedom in a limited space is a huge technical challenge for dexterous hand hardware.

In addition to the hardware challenges, the hands must also interact with the environment to achieve dexterity. When paying attention to the geometric characteristics of objects such as shape, size, and position, it is inseparable from the core sensing component of the humanoid robot, the sensor, to conduct real-time multi-dimensional perception of the operating environment and objects. However, at present, in addition to the more common applications of visual sensors in the industry, and the landing of tactile sensors, the industry still lacks multi-dimensional sensors integrating vision, touch, hearing and force.

To be versatile enough, in addition to the geometric properties, you should also pay attention to the semantic properties of the object, such as what the object is and what it is used for. Only in this way can humanoid robots interact effectively in unknown environments and achieve truly dexterous operation.

Compared with the dexterity of the upper limbs and the dexterity of the hands, the ability to move the lower limbs seems to progress faster in terms of technology. This is probably due to the combination of software and hardware and the development of artificial intelligence technology, especially reinforcement learning.

The CL-1, which was tested for extreme impact at the exhibition site, demonstrated strong dynamic balancing capabilities. In its earlier demonstration video, the CL-1 can also hike in the mountains, descending gravel slopes, climbing grassy slopes, and walking ditches. Founder Zhang Wei once said, "The leg, that is, the lower limb, does not require new physical discoveries, but it is a process of engineering iteration that requires continuous experimentation." "This probably explains why the humanoid robot's lower limb capabilities are progressing faster.

The lower limbs of humanoid robots are divided into wheeled, quadrupedal, and bipedal. Among them, quadruped robots focus on solving the "general mobility capability", that is, to achieve generalized adaptation to various terrains. Bipedal robots are designed to walk upright so that the front two limbs (i.e., both arms) can be operated.

The stability and balance of both feet are twice as difficult as those of four legs. That's why we can see four-legged dexterous dogs, climbing up and down stairs super silky, while most bipedal humanoid robots still tremble when they walk.

What is certain, however, is that from wheeled to quadrupedal to bipedal, the technical difficulty and cost are incremental. From the point of view of cost, the wheel-foot solution is almost a fraction of the bipedal. Different options represent different technical difficulties and different costs, and also affect the commercialization process of humanoid robot companies.

In general, humanoid robots are essentially moving or moving or operating instead of people. Some robots focus on movement, while others focus on operation. Therefore, we will see that most companies that focus on the development of dexterous hands have wheeled lower bodies, while bipedal robots have more upper body ends with grippers.

First industry, then commerce, and finally into the home

After talking about product configuration, let's talk about the landing scenario that everyone is more concerned about. At present, the lack of mature and rigid scenarios is one of the biggest problems faced by humanoid general-purpose robots.

According to the research of the Tiger Sniff Think Tank, industrial manufacturing, scientific research and innovation, business services, and home services are the four fields that are more likely to apply humanoid robots at present.

Further, those scenarios in the four fields that are relatively closed, have relatively single capacity requirements, higher labor costs, or higher risk factors will be implemented faster, such as handling and quality inspection in the automotive industry production line, supermarket replacement, replenishment and other scenarios.

Regarding the order of landing in different scenes, the more general view is that industry first, then commerce, and then into the home.

From the perspective of ability and willingness to pay, industrial scenarios are indeed more likely to be the first to land. On the one hand, because the automation level of factories is higher, and manufacturers have clear needs in some scenarios, they have more ability to pay.

For example, UBTECH focused on key manufacturing areas such as automobiles and 3C for the first time at the meeting and launched the "Humanoid Robot Industrial Scenario Solution". Its Walker S series humanoid robots can be oriented to multi-task industrial scenarios, completing a variety of tasks such as intelligent handling, intelligent sorting, intelligent quality inspection, screw tightening, parts installation, and process material operation.

When talking about why it focuses on industrial scenarios, Dr. Qin Wenlong, an expert in humanoid robot technology at UBTECH, said that compared with service scenarios and home scenarios, industrial scenarios are relatively structured, and the factory environment and task content are determined. In addition to structuring, the industrial production line has a more urgent demand for automation and intelligent upgrades, and new forms of high-end intelligent manufacturing equipment are urgently needed to make up for the blank areas outside the task execution scope of existing industrial robots, collaborative robots, composite robots and other intelligent equipment.

In addition to industry, when this wave of embodied intelligent robot companies are looking for a commercialization path, many of them are also the first to cut into the home scene.

Because for the new embodied intelligent robots, compared with the accuracy and speed of traditional special robots, it is more feasible to find some scenarios with high replacement value in terms of versatility, such as home care scenarios such as elderly care and taking children. In these scenarios, more emphasis is placed on interaction with people and service to people, rather than precision.

In addition, because of the family scene, the market size is huge, and the ceiling is higher, but it is also more difficult to implement the technology.

Ju Xiaozhu said, "At the beginning, everyone may think about the family scene first. But in fact, after doing it, I found that the needs of every family and each person are strange, too random and unstructured, and the generalization and intelligence of humanoid robots are very high, and the landing will not be fast. "

In addition to the non-structural environmental characteristics of the family, the technical maturity of the robot itself, the safety, and the acceptance of humanoid robots by family members are all problems to be solved for humanoid robots to land in the family.
In some business service scenarios, such as marketing services, front desk reception, etc., dedicated traditional robots have been used for a long time. It seems that at present, compared with the humanoid and not the humanoid, everyone is competing more to understand the needs of the enterprise in this scenario. At this point, it is clear that commercial robot players who have already entered the game have an advantage.

In addition, for the scientific research and education scenario, there are currently a number of humanoid robot companies entering the game, and the competition is fierce, but the ceiling of this scenario is relatively low, and the mass production scale does not meet market expectations, so it is more like everyone's short-term helplessness.

Regarding the specific landing time in each scenario, in the survey of the Tiger Sniff Think Tank, the consensus of most manufacturers is that in 1-3 years, humanoid robots will usher in the landing in the industrial scene, 3-5 years in the commercial service scene, and 5-10 years in the family scene.

The Tiger Sniff think tank believes that this judgment may still be too optimistic. Because the progress of commercialization of new technologies is generally slow, the premise of commercialization is that it is necessary to find a very clear product demand scenario. Although UBTECH and other enterprises have entered the stage of factory training in some scenarios such as automobile manufacturing, the market demand of relevant scenarios needs to be further verified in consideration of solving the problem of versatility and cost.

Perhaps it seems to be a feasible way to commercialize while trying to maintain the height of technology research and development, gradually dismantling the technology, reducing the dimension to find some scenarios that can be implemented, exploring the direction of productization, and solving the problem of universality, but it is not easy.

Moreover, if it is to be successfully commercialized, the product must also achieve at least 80 points and be able to open a qualitative gap with competitors. But at present, the differentiation of everyone's products is not obvious in terms of appearance or function.

Overall, at present, although humanoid robots are very popular, the technical route and business scenarios are still not clear, there is still a long way to go before mass production, and commercialization is still a long way off. It's still too early to talk about embodied intelligence ushering in the iPhone moment.

This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com