[Edaily Reporter Kim Hyun-ah ] AI semiconductor company Mobilint is developing an artificial intelligence (AI) computing platform for humanoid robots. This technology enables multiple AI systems to run simultaneously within the robot and translates analysis results into actual movements. The company has focused on integrated development—from AI semiconductors to onboard modules and software—to reduce power consumption and processing latency. It plans to verify the platform’s applicability by integrating it with LGELECTRONICS’ humanoid system environment in the future.

Mobilint announced that it is designing a neural processing unit (NPU) and AI acceleration chips specifically tailored for humanoid robots, and is simultaneously developing modules for robot integration as well as software to support the conversion, optimization, and execution of AI models.

Multiple AI Systems Collaborate from Perception to Action

For a humanoid robot to move autonomously in the same space as humans, AI functions—such as perceiving the surrounding environment, understanding human commands, and determining the next action—must operate simultaneously. This requires a wide range of computations, from processing camera and sensor data to setting movement paths, making safety judgments, and controlling joints.

Vision-Language Models (VLM), which understand both visuals and language, and Vision-Language-Action Models (VLA), which translate decisions into actual actions, are also utilized. If computations are delayed while multiple AI systems operate simultaneously, this can affect the robot’s responses and movements.

Mobilint plans to optimize both the NPU and software together to efficiently process multiple AI tasks and ensure that necessary computations are performed in a timely manner.

Low-Power AI Execution Inside the Robot

Unlike data centers, humanoid robots operate on batteries within a limited space. The ability to manage power consumption and heat generation is just as important as high computational performance. To reduce dependence on external servers or communication networks, AI must be able to run stably within the robot itself.

Mobilin is developing high-performance, low-power NPUs and compact accelerator modules, while also establishing a software development environment that adapts AI models to real-world robotic environments. The goal is to reduce the burden on robot manufacturers, who would otherwise have to separately integrate and optimize semiconductors, models, hardware, and software.

Performance Validation in LGELECTRONICS’ Humanoid Environment

The developed platform will be validated in conjunction with the “ LGELECTRONICS(066570)” environment for humanoid systems in everyday spaces. The company plans to verify how quickly and stably the process—from sensor input to recognition, decision-making, and actual robot control—proceeds when multiple AI systems operate simultaneously.

Power consumption, heat generation, and long-term operational stability under battery-powered conditions will also be examined. The reliability and heat dissipation performance of the accelerometer modules will be assessed, taking into account the vibrations and shocks generated by the robot’s walking and joint movements. Any processing delays, power consumption, or heat generation issues identified during the verification process will be addressed through improvements to the semiconductors, modules, and software.

This development is part of the Ministry of Trade, Industry and Energy’s “K-On-Device AI Semiconductor Technology Development Project.” Mobilint serves as the lead research and development institution for the “Development of On-Device AI Accelerators for Humanoids” project. The company plans to apply the developed technology to service robots, autonomous mobile robots (AMRs), smart home appliances, and AI cameras in the future.

Park Ju-hyun, Head of Research Planning at Mobilint, stated, “In physical AI, it is not only computational performance that matters, but also whether the AI’s decisions are rapidly translated into the machine’s movements.” He added, “We will conduct integrated development of semiconductors, modules, and software, and verify their applicability in LGELECTRONICS’ humanoid environment.”