news

interface news reporter | lu keyan

interface news editor | wen shuqi

after the closing of the world robot conference 2024, the discussion on humanoid robots has changed again. while it has attracted countless investors, there are also many voices that believe that humanoid robots are far from flexible enough, and their seemingly intelligent performance is limited to demos, and commercialization is a distant thing.

at present, humanoid robots have formed a commercial closed loop to a certain extent. for example, the scientific research industry is willing to pay a high price to purchase them for auxiliary research and development. however, due to the level of intelligence and the difficulty of mass production, such products can only be shipped in small batches, and it is difficult for humanoid robot companies to make money.

but this does not stop the industrial sector's enthusiasm for humanoid robots, especially in the highly automated automobile manufacturing sector.

ubtech is the first robot manufacturer in china and even in the world to implement humanoid robot training in factories, and has received more than 500 orders so far. this year, ubtech has reached cooperation with dongfeng liuzhou motor, geely automobile, faw hongqi, faw-volkswagen qingdao branch, audi faw, nio and other automobile manufacturers, and its industrial version of humanoid robot walker s series has entered the production line training.

"when can you start production?" this is the most common question that car manufacturers ask ubtech, in addition to reliability and price. one background is that china's manufacturing industry is facing a labor shortage of more than 30 million people, and the high employee turnover rate has become one of the most troublesome problems in the manufacturing industry.

ubtech robotics chief brand officer tan min mentioned that the annual worker turnover rate at a car factory with nearly 1 million workers is around 30%. adding the salary and training costs of each worker, the factory's demand for robots can be imagined.

in fact, automobile manufacturers have widely used automated equipment such as robotic arms in the production process, but this is still not enough.

vice president of ubtech robotics and executive director of the research institutejiao jichao said that compared with humanoid robots, industrial robotic arms have limitations in space and application: they are usually fixed and can only operate in a limited area. the workflow is pre-set. faced with special circumstances or changes that may arise in the production process, such as mixed-line production or mixed loading, product size and appearance may change frequently, and robotic arms cannot adapt.

flexibility is one of the important requirements of the manufacturing industry, and the advantage of humanoid robots lies in their high flexibility and freedom. the human body and hands have more than 40 degrees of freedom, and even without the hands there are 28 degrees of freedom, which enables humanoid robots to perform more complex and diverse movements, such as bending over, lifting legs, etc.

and because modern automotive factories are typically designed around the anatomy and abilities of human workers, humanoid robots are better able to adapt to these environments without requiring a wholesale overhaul of the factory layout.

global automakers have the same needs. bmw chose figure, a well-known american humanoid robot company, mercedes chose another startup apptronik, and tesla also announced that it will deploy 1,000 humanoid robots optimus to its car factories next year to perform tasks such as sorting battery cells.

however, the progress of humanoid robots entering factories at home and abroad is roughly the same. they must be trained on the production line for 12-24 months. this process is also called concept testing. car manufacturers need to use this to verify whether humanoid robots can really improve efficiency. this is also the most important step in their commercialization process.

what can humanoid robots do in factories? take ubtech's walker s as an example. it can tighten bolts, transfer parts, install parts, attach vehicle logos, and perform other tasks. it can also use real-time image capture and transmission functions to complete quality inspections of vehicle door locks, seat belts, lampshades, interior decoration, and other parts.

jiao jichao told jiemian news that the demands that car manufacturers put forward to robot manufacturers are basically based on the existing capabilities of humanoid robots, and they are willing to give emerging products a certain amount of development space. in addition to buying robots, car manufacturers will also provide application scenarios to ubtech to help optimize the latter's algorithm capabilities.

another significance of practical training is to allow robots to obtain more data from the physical world, thereby improving their intelligence level. jiao jichao believes that the more car manufacturers they cooperate with, the more data on common scenarios they can capture, making the data more generalizable and having a large number of flexible small models. compared with general large models, these small models may have more advantages in specific applications and are also more cost-effective.

at present, the industry does not need a universal robot. ubtech's idea is that with small models for specific fields, robot manufacturers can train robots to do different jobs by adjusting the software program and configuring them according to needs, without frequently changing the hardware. in some cases, it may only require replacing the actuator, such as replacing the hand actuator with a two-finger type.

this is a beautiful idea, but robot manufacturers still need to solve the problem of mass production, which is also a common challenge for humanoid robot companies around the world.it was originally expected that optimus would be put into factory operation by the end of 2024 and would be available for sale at the end of 2025 at the earliest, but the planned time for large-scale mass production has now been postponed from 2025 to 2016.

but the supply chain is working hard. jiao jichao mentioned that the high-power servo drives currently used in ubtech's humanoid robots involve about 50 to 100 parts, including motors, reducers, encoders, mcus, torque sensors, etc.localization rateabout 90%; the number of parts of a small power servo drive is about 40-80.localization ratemore than 95%.

where there is demand, there is a market, and industrial manufacturing is considered to be the most likely direction for the large-scale commercialization of humanoid robots. before that, robot manufacturers still need to continue to iterate hardware capabilities to enable robots to have stable and reliable movement capabilities in multi-dimensional space, enable them to undertake more complex tasks, and continuously improve their intelligence level.

a consensus among car manufacturers and robot manufacturers is that when humanoid robots can truly solve factory problems, cost will not be an important issue, because demand will inevitably increase, and supply chain capabilities will also improve, just like the process from the advent of cars and mobile phones to their popularization. for that possible moment of explosion, they are willing to wait.