German car company BMW is accelerating the application of humanoid robots in automobile production. Starting this summer, BMW’s factory in Leipzig, Germany, will officially deploy two humanoid robots Aeon developed by Hexagon Robotics for the first time in Europe for use in the car manufacturing process. Currently, the robot is being tested in the factory.

Michael Nikolaides, head of production process management and digitalization at BMW, said: "This will be the development direction of future automobile manufacturing."

Adapt to existing production lines

For a long time, the automobile industry has widely used automated equipment such as robotic arms, but the core reason why BMW chose humanoid robots this time is that they can directly adapt to the existing production system without the need for large-scale transformation of production lines.

Nikolaides pointed out that humanoid robots have similar body shapes and operational abilities to humans, so they can directly enter jobs currently performed by workers. The cost for companies to redesign production lines around robots is very high. Nowadays, the cost of robots has dropped significantly, and a more reasonable way is to adapt robots to existing workflows.

Bill Ray, vice president analyst at market research firm Gartner, also believes that as the price of robots drops, there is no need to redesign factories for robots.

The Aeon robot put into testing this time is about 1.65 meters tall, weighs 60 kilograms, has a maximum moving speed of 2.4 meters per second, can carry 15 kilograms of objects in a short time, and has a continuous load capacity of about 8 kilograms. The robot is equipped with 21 sensors, including cameras, radars, microphones, and torque sensors, for environmental perception and precise operation.

In terms of robot shape design, Aeon does not use a bipedal structure, but uses a wheeled mobile chassis. BMW believes that in a flat factory floor environment, the wheeled design is more efficient and stable than the two-legged design.

However, the robot's endurance is still a shortcoming. Aeon's single battery life is about 3 hours, while the factory standard shift is usually 8 hours. To this end, Hexagon has designed an automatic battery swapping system for the robot. The robot can go to the battery swapping station on its own and complete the battery replacement in about 3 minutes.

According to BMW's plan, Aeon will initially be responsible for repetitive tasks such as parts transportation and grabbing and placing in battery assembly. Although robots have multiple capabilities, similar to human workers, their work content does not switch frequently.

Nikolaides believes that humanoid robots can not only take on highly repetitive and high-intensity jobs in the future, but can also alleviate the increasingly prominent labor shortage problem in the manufacturing industry.

“Manufacturing labor shortages will become increasingly evident in the coming years, and humanoid robots can help.”

However, Nikolaides also emphasized that robots will not simply replace human employment.

"When automation emerged in the automotive industry in the 1970s, many people were worried that jobs would disappear, but in the end it gave rise to a large number of new occupations. We also hold the same view on humanoid robots."

Advanced training technology

During the training process, BMW combined "digital twin" factory simulation and remote control technology. With the help of NVIDIA's software system, engineers first repeatedly simulate tasks in a virtual factory, allowing the robot to find the optimal action path through reinforcement learning. At the same time, the robot will also learn the movements of human workers to complete operations such as grabbing and carrying.

Currently, robot training technology is evolving rapidly. Arnaud Robert, president of Hexagon's robotics business, said that "imitation learning" is one of the most promising directions in the field of Physical AI. Robots can learn how humans complete a task through multi-angle videos or human motion sensors, thereby reducing training time from months to days.

Robert said that the ultimate ideal state is for robots to immediately learn to do the same tasks after observing humans working. "This goal may be achieved in the next one to two years."

However, humanoid robots are still far away from fully autonomous operations. Gartner predicts that within the next three to five years, robots will be able to truly complete industrial tasks independently through simple voice commands.

Car companies intensively deploy humanoid robots

In addition to BMW, many car companies are also deploying humanoid robots. Japanese car company Toyota plans to introduce robots from Agility Robotics after a successful pilot. Xiaomi has already tested its self-developed humanoid robot in the production process of electric vehicles.

At the same time, South Korean car company Hyundai is using the four-legged robot Spot developed by Boston Dynamics to conduct industrial inspections, and plans to introduce the Atlas humanoid robot in the future.

In fact, BMW has previously tested the Figure O2 robot developed by Figure at its Spartanburg plant in the United States. This robot has been involved in the production of approximately 30,000 X3 models, and its work rhythm is basically the same as that of human workers.

BMW found in practical applications that compared to traditional industrial robots, AI-driven humanoid robots perform better in coping with complex environmental changes. For example, when there is a slight deviation in the position of a metal part, a traditional robot may stop due to errors, but an AI robot can analyze it in real time and continue to complete the task.

In addition to humanoid robots, BMW has also previously used Spot robots to perform factory inspection tasks, including entering underground equipment areas, going up and down stairs, etc.

Regarding the robots entering the factory, BMW said employees have responded positively overall. The company predicts that in the future, employees will even come up with names for these robots, just like they do with traditional industrial robots. Ray believes that once people give a robot a name, they will regard it as a "collaboration object" rather than a simple machine, and will be more tolerant of its mistakes.

However, Ray also reminded that there is a certain degree of over-hype for humanoid robots in the current market. "Many robot demonstrations today are more like capital market performances, used to push up company valuations." People may overestimate the capabilities of robots. He explained:

"When you see a robot walking, you assume it can run, climb, and jump. It may not be able to do that, but your brain has filled in the gaps. We always have unrealistic expectations when people show robots off."