Humanoid Robots Work in a Chinese Factory

Robotics startup MagicLab introduces a fleet of humanoid robots at a factory to train them in a variety of jobs .

MagicBot robots work together in a factory. (Video: MagicLab).

MagicLab's MagicBot performs a variety of tasks, including production monitoring, material handling, parts picking and sorting, inventory checking and management. The humanoid robots also demonstrated their ability to operate collaboratively on a small scale, proving that they can work effectively together, Interesting Engineering reported on December 9.

In January, MagicLab unveiled its third-generation humanoid robot, MagicBot, designed to compete with Boston Dynamics' Atlas robot. In February, it shared videos of the robot roasting marshmallows, performing magic tricks, and dancing. MagicLab joins a wave of companies deploying robots in industrial settings as China aims to mass-produce humanoid robots by 2025 and dominate the market by 2027. Founded in December 2023, MagicLab (Magic Atom Robotics Technology) is focused on revolutionizing research and rescue, logistics, and manufacturing with humanoid robots.

In the new video, several MagicBot humanoid robots work together. One robot passes a tray of materials to another robot at a workstation after lifting it from an assembly point. Their effective collaboration is demonstrated as the second robot takes items out of a box and transfers them to a processing machine.

"Compared with traditional composite robots, humanoid robots have higher adaptability to complex environments due to their human-like characteristics and can be adjusted to different jobs and production line layouts. It is especially suitable for the fields of computers, communications, consumer electronics, home appliances and many other factories that need to frequently change production lines," said Wu Changzheng, director of Magic Atom.

Currently, the humanoid robot MagicBot used in the factory can rotate freely 42 degrees. Its arms can lift up to 20 kg while its entire body can transport 40 kg of goods. Made from ultra-light, high-strength materials for increased efficiency, MagicBot has a battery life of up to 5 hours when walking and working continuously. In addition to roasting marshmallows on a tray, folding children's clothes and placing them in a basket, the remote-controlled robot can also water plants and dance to lively music. It also possesses impressive dexterity when grasping small objects and performing various tasks with smooth agility.

The MagicBot humanoid robot used in the factory can rotate 42 degrees of freedom.

MagicBot achieves these advanced skills by using miniature high-torque servos combined with highly sensitive multi-directional pressure sensors, allowing for exceptionally smooth hand movements. The developers emphasize that the six servos play a key role in allowing it to manipulate flexible objects, ensuring users can control them with precision and intuitiveness. The robot's hands excel at grasping objects of various shapes, sizes, weights, and materials, including soft, smooth, or pliable objects.

According to its creators, MagicBot can simulate about 70% of human hand movements, meeting the requirements for a wide range of applications, from household chores and public services to specialized industrial operations and hazardous material handling. MagicBot is also equipped with many sensors. Sensor data is synthesized by an algorithm to create a hyper-aware sensor, allowing the robot to perceive the environment 360 degrees and recognize semantics.

According to Sango Automation, the cognitive capabilities enable MagicBot to plan a safe path, avoid obstacles, and reach its operating point efficiently using a self-developed navigation algorithm. Through augmented data training, MagicBot continuously improves its ability to perceive the environment, make decisions, and execute tasks. MagicBot uses four types of training data: synthetic data, remote operation data, imitation learning data, and real-time field data. To optimize the use of data, the team developed the MagicData AI engine to annotate data, process, and train the simulation.

The new MagicBot robot, expected to launch in early 2025, will enter small-scale production, and can be deployed in industrial and commercial sectors.