AI-powered humanoid robots move from labs to real-world jobs

With the rapid advancement of artificial intelligence and robotics, many robot manufacturers are transitioning from prototypes to mass production and commercial deployment this year.

The year 2026 has marked a turning point as humanoid robots move beyond research laboratories and begin entering real workplaces. Industry experts describe this shift as moving from the stage of "demonstrating what robots can do" to "proving what they can actually accomplish."

Around 13,000 humanoid robots were shipped to companies worldwide last year. According to Goldman Sachs, that figure is expected to reach between 50,000 and 100,000 units in 2026, CE Report quotes Anadolu Agency.

Humanoid robots are already being deployed in a wide range of environments, including airports, automotive factories and logistics warehouses. Experts expect 2028-2030 to be the period when humanoid robots become a common sight in everyday life.

Among the companies developing humanoid robots is Apptronik, a Texas-based startup founded by two University of Texas academics who advised NASA on humanoid robotics projects, together with an entrepreneur. At the end of June, the company opened Robot Park, an 8,000-square-meter facility described as a "robot school," where hundreds of humanoid robots are trained simultaneously.

Robots trained around the clock

Apptronik introduced its first humanoid robot, Apollo 1, in 2023 and unveiled Apollo 2, which has been undergoing training since February 2025, in June this year.

At Robot Park, hundreds of Apollo 2 robots undergo continuous training 24 hours a day, seven days a week, learning skills they will need in real-world workplaces.

The robots repeatedly perform tasks they are expected to encounter at customer sites, including loading boxes onto conveyor belts, sorting toys and vehicle parts into bins, stocking shelves, packing goods, moving pallets, scanning inventory and opening doors.

They also practice walking on uneven surfaces, grasping and releasing objects, and navigating spaces designed for humans.

The facility trains two versions of Apollo 2 simultaneously. The wheeled version is intended for high-efficiency industrial operations and complies with existing industrial mobile robot safety standards, while the bipedal version focuses on improving walking performance and reliability in complex environments designed for people.

Learning by imitating humans

Much of the training is carried out through teleoperation. Human operators stand alongside the robots and control their movements in real time through a control interface.

The robot imitates the operator's actions while sensor data, camera images, joint positions and force measurements are recorded continuously.

These data are used to train the robot's AI system, which is integrated with Google DeepMind's Gemini Robotics models. Every action performed by the operator becomes a training example, helping the robot learn how to respond when it encounters similar situations in the future.

As the AI model improves, the robot requires less human guidance and becomes increasingly autonomous. Company officials describe the process as a "continuous learning loop," in which robots perform tasks, collect data and steadily improve with each training cycle in real-world environments.

The knowledge and technological advances gained through this process are expected to pave the way for the next generation of humanoid robots. Initially designed to work alongside people in manufacturing and logistics, they are also being developed for future use in healthcare, homes and many other sectors.

Photo: Chat GPT