Dialogue with Lingbao CASBOT: Humanoid Robots Entering Factories Start with Learning | Exclusive Interview by 36Kr

Text | Wang Fangyu

Editor | Su Jianxun

Currently, embodied intelligence is undoubtedly one of the hottest tracks.

Data from GGII shows that from January to October this year, the global financing amount in the humanoid robot field has exceeded 11 billion yuan, far exceeding the whole year of 2023. Behind each booming financing, entrepreneurs are collectively sprinting towards the dream of a general intelligent robot.

The large model is a key driving factor for the upsurge of humanoid robots. The emergence of the large model capability has transformed robots from feature phones to smart phones, enabling them to learn and adapt to the environment, and also injecting new hope into the general intelligent robot.

Since this year, many domestic humanoid robot companies have released their first humanoid robot products and are actively exploring downstream application scenarios. And the automotive factory has become the first choice for the landing of humanoid robots - UBTECH robots have entered the factories of NIO and Zeekr, and Fourier robots have started working in the SAIC-GM factory... Some investors joked: "Recently, every robot is being sent to the car factories, and the car factories are almost not enough."

Unlike the above-mentioned humanoid robot manufacturers, Lingbao CASBOT has chosen the first landing factory to be the intelligent manufacturing factory of Lenovo Group, mainly responsible for the more precise operation of notebook assembly.

Lingbao CASBOT is a humanoid robot brand under Beijing Zhongke Huiling Robot Technology Co., Ltd., dedicated to researching and developing general humanoid robots and embodied intelligent products for scene landing. On November 13, Lingbao CASBOT released its first full-size biped humanoid robot CASBOT01 - a multi-scene landing general brain-like intelligent robot.

After the conference, Lingbao CASBOT co-founder and COO Zhang Miao and CTO Ma Shikui communicated with 36Kr and other media. Behind the financing fever and the factory-entry fever, at what stage has the humanoid robot industry developed? How far is it from the truly general robot? Lingbao CASBOT gave its own judgment and thinking.

The following is the communication record between 36Kr and Lingbao CASBOT (edited)

The intelligent level of humanoid robots is currently between L3 and L4.

Media: What is the current development status of the intelligence level, operational ability, and cost of humanoid robots, and at what stage of development are they?

Ma Shikui: For the intelligence level of humanoid robots, some standards are gradually emerging. Most of them still refer to the classification method of L1 to L5 of Tesla's autonomous driving. According to this standard, the current humanoid robots have not reached the L4 level (that is, the completely autonomous or minimally monitored state), and are basically in a transitional state between L3 and L4.

The human-computer interaction language understanding ability of the large model is very strong, and it is also relatively mature in the visual aspect; in the motion control aspect, we have seen some underlying capabilities of the cerebellum manifested from different gene manufacturers, and this ability is relatively mature, but it is basically still at the demonstration stage; however, at present, some operations are relatively mature and can be directly commercialized, such as grasping, placing, plugging, and unplugging.

The cost will have further room for decline with the scale development, and in terms of price, the cost is decreasing very rapidly from the industry trend.

Media: From a quantitative perspective, how to quantitatively define the small-scale mass production and large-scale mass production of humanoid robots?

Zhang Miao: Small-scale and large-scale are relative concepts, and each company may have its own definition. We think that in the current situation, within 100 sets is considered a small-scale level, and more than 100 sets is considered a relatively large scale.

Media: In the field of humanoid robots, what obvious development trends will there be in the next one to three years?

Zhang Miao: First, we have seen that from the policy level, since the second half of 2023, the national level has issued many policies to support the development of the humanoid robot industry. Many provinces are now also beginning to follow the footsteps of the country and successively promulgate and issue policies related to the humanoid robot industry, and there will be more industrial policies to support the development of humanoid robots.

Second, with the previous policy support, more component and upstream material enterprises will follow the development of the main engine factory to further improve and drive the supply chain to further mature.

Third is commercial application. It is expected that many embodied intelligent and humanoid robot manufacturers will cooperate or jointly develop with various fields and application scenarios to promote the application of embodied robots in various fields.

Finally, if the embodied robot is to move towards the C-end scene in the future, then some issues in terms of safety and ethics will gradually be brought to the table for discussion, and this may also be a trend.

Media: Lingbao CASBOT humanoid robots are mainly for the B-end. How do you view the overall progress of the industry for humanoid robots entering the C-end household market?

Zhang Miao: Before 2024, humanoid robots or embodied intelligence were basically in a stage of platform building, and everyone was actually completing the R & D from 0 to 1. After having the basic platform, there is still a relatively large distance from the product and commercialization, especially when combined with various application scenarios, and the performance needs to be improved on the basis of the scenarios. I think this cycle will last for a longer time.

In the C-end household service scenario, based on our past experience, it is relatively sensitive to cost. The cost of the biped robot is still quite a gap compared to the expectations of C-end customers.

Lingbao CASBOT's strategic plan is to continuously polish its product capabilities through real scenarios, simultaneously reserve on the R & D side for both the B-end and C-end, and choose a more appropriate time to move towards the C-end consumer market at the commercialization level according to the internal and external situation of the company.

Ma Shikui: Musk's judgment on humanoid robots can be used as a reference. He expects that in 2025, humanoid robots will work in Tesla factories to replace some human work, and in 2026, they will be open to third parties for services, and gradually finally reach the C-end household scene. So the household scene is still more difficult.

Humanoid robots, learning in application scenarios

Media: Currently, many manufacturing workshops still use mechanical arms. What are the differences between mechanical arms and humanoid robots in factory applications?

Ma Shikui: There are two differences. One is the chassis. The chassis of a humanoid robot is movable, more flexible, and applicable to multiple scenarios, while the traditional mechanical arm is fixed and its movement is more limited. The other is intelligence. Humanoid robots are more versatile and can quickly learn various operations and skills.

Taking the notebook assembly process as an example, the material box of the notebook has a fixed shape, but the placement of the material box will be in different positions, and the hole positions of different notebooks are also different. It is very difficult for the traditional mechanical arm to learn, while the humanoid robot has an advantage. The advantage of the traditional arm is that it is simple, repetitive, efficient, and low-cost. This type of work is still done by the traditional mechanical arm. What the traditional mechanical arm cannot do can be replaced by the humanoid robot in the future. The two are a good complement and can collaborate to complete some tasks.

Media: How many core components of Lingbao CASBOT are self-developed, and what considerations are there for integration and self-development?

Ma Shikui: The advantage of self-development is that it is autonomous and controllable, and can make a difference, but the relative upfront investment cost is high, the product launch progress is slow, and there may also be some uncertain risks.

All the software of our humanoid robot is self-developed, and not 100% of the hardware is self-developed. Some chips and sensors are purchased externally. We do not pursue full self-development, but instead, we infer from the scene requirements to decide whether to self-develop.

For example, our dexterous hand is self-developed because we found that the current general dexterous hand products on the market cannot meet the requirements of the factory assembly scene for refinement.

Media: The company has chosen the assembly link of the Lenovo notebook factory as the landing scene. What operations do humanoid robots need to perform in this link, how difficult is it, and what are the differences compared to the automotive factory scene?

Ma Shikui: Lingbao CASBOT chose the manufacturing scene and started with notebook assembly. One reason is that Lenovo is our shareholder, making it easy to enter; more importantly, we have accumulated a lot in terms of precise micro-assembly and flexible operations, which can make this (humanoid robot assembly) a reality in the first place.

For the automotive factory, (the humanoid robot) is more about moving large items, and the requirements for the load will be higher. The operations of notebook assembly include complex operations such as grasping, twisting, plugging, unplugging, and using tools, and the requirements for precision will be very high. There are also requirements for flexibility, and the generalization requirements are also raised. Correspondingly, the notebook assembly has higher requirements for the ability of the hand and the corresponding AI capabilities.

For humanoid robots, we believe that with two arms and dexterous hands, the collaboration method is empowered by the brain, and it can be regarded as humanoid. So the robot in the assembly scene is actually not bipedal, only the upper body of a person.

Media: Whether it is screwing or carrying in the factory, it seems that the original industrial robots can achieve higher efficiency. Now that humanoid robots are doing these procedural things, is it going against the trend?

Ma Shikui: From the perspective of the future of humanoid robots, it has two main advantages. The first is versatility, that is, a humanoid robot can do many kinds of work; the second is cost, and the cost must be reduced to a certain level, otherwise it cannot compete with specialized robots.

Then why are humanoid robots entering the factory now? Because to achieve versatility, a sufficient amount of data must be accumulated. On the one hand, data accumulation relies on simulation, and on the other hand, it must also be accumulated in specific scenarios. Moreover, in the semi-structured factory scene, there is no need to worry about ethical issues as in the C-end scene.

In addition, humanoid robots entering the assembly factory will not directly compete with traditional mechanical arms. They choose things that mechanical arms cannot do. For example, many processes of notebook assembly require both hands to collaborate to complete; in addition, the dexterous hands used to be two-finger or three-finger, while humanoid robots require high-degree-of-freedom simulation hands, so that some fine, force-sensing, and compliant controls can be achieved.

Zhang Miao: Traditional industrial robots are very fast, and their efficiency and precision are also very high, but they are too customized. Many of them are very large mechanical devices with fixed processes and do not directly interact with people.

At present, many domestic factory production lines are developing in the direction of flexible manufacturing, and customized production needs to be done according to customer needs. Then traditional industrial robots cannot be used on such production lines. And the embodied robot is more adaptable to the trend of flexible manufacturing.

Lingbao CASBOT is exploring in the assembly scene of Lenovo. Although the current development cannot fully meet the needs of the production line, it has a process of gradual realization and progress, and we are in the process now.

Media: Speaking of assembly, in the future, can labor-intensive 3C manufacturing enterprises like Foxconn apply robots on a large scale to replace humans in assembling mobile phones?

Ma Shikui: It is actually more difficult for humanoid robots to enter the mobile phone production line than the notebook. Because the mobile phone is smaller and more precise. Foxconn has been exploring this matter and has also established a robot business department. I think the general trend is unstoppable.