Kia develops AI-based robot control technology... Advancing manufacturing automation with 'Imitation Learning and Simulation'

Kia is focusing on developing an "intelligent control brain" that allows robots to judge and move autonomously using artificial intelligence (AI). This is a strategy to overcome the limitations of existing methods, where robot movements had to be individually modified every time a vehicle model or process changed, and to implement flexible automation at manufacturing sites.

Transitioning from coordinate input methods to 'Imitation Learning and Simulation' technology

According to a video from HYUNDAI, the manufacturing SI (System Integration) technology development role at Kia is advancing robot control technology by utilizing reinforcement learning and imitation learning technologies. The existing robot teaching method was a passive approach where workers used a motion device called a 'teach pendant' to individually specify and save the coordinate values of robot joints. However, there was the inconvenience of having to teach new tasks every time the process environment changed.

To solve this, Kia is securing robot control technology centered on four core technologies. First, 'imitation learning,' which transfers an expert's work know-how to the robot, is key. Through teleoperation using a data glove, human intuitive movements are directly taught to the robot in the form of data. Additionally, 'precision motion tracking' technology is utilized, which extracts human position and rotation data using a tracker attached to the back of the hand to control the trajectory of the robot arm.

In particular, to overcome the difference in physical structure between humans and robots, 'real-time retargeting' technology, which converts human movements to robot specifications in real time, and 'simulation (Sim-to-Real)' technology, which transplants control models verified in a virtual environment directly into actual robots, are treated as important. The person appearing in the video explained the technology as "the process where intelligence learned in a virtual environment moves into reality."

The intersection of software and hardware... Aiming to advance 'Physical AI'

Kia's manufacturing SI technology development role is an area where software and hardware intersect, and an understanding of robotics (kinematics and dynamics) along with the ability to implement deep learning and reinforcement learning algorithms is essentially required. The participant cited 'problem-solving ability' and 'persistence'—analyzing the causes of learning failures and redesigning algorithms—as core competencies.

At the organizational level, to increase development efficiency, actual robot hardware infrastructure is provided so that learned algorithms can be tested immediately, and the use of AI coding assistant tools is actively supported. The participant stated, "The charm lies in the fact that the written code and algorithms do not stay only inside the monitor, but make robots in the actual physical world move intelligently." Regarding future goals, they said, "It is to advance the recently highlighted robotics and Physical AI technologies to complete robot control technology that acts flexibly like a human in various manufacturing environments."