Press Release
September 8, 2026
.png)
Nota AI, a company specializing in AI model compression and optimization, announced that it will work with LG Electronics and Mobilint to develop humanoid robots powered by Korean-developed on-device AI processors. Nota AI will optimize large-scale robotics AI models for Mobilint’s neural processing unit (NPU), enabling them to run efficiently in resource-constrained robot environments.
Nota AI has been selected to lead the humanoid AI model optimization track of Korea’s K-On-Device AI Semiconductor Technology Development Program, an initiative led by the Ministry of Trade, Industry and Energy and the Korea Evaluation Institute of Industrial Technology (KEIT).
The project marks a significant expansion of Nota AI’s optimization technology into physical AI, building on its experience across mobile, edge and data center environments. Nota AI will lead a dedicated workstream focused on efficiently connecting AI models, NPUs and robotic systems — an increasingly critical capability for deploying advanced AI directly on physical devices.
The broader initiative aims to develop autonomous humanoid robots capable of perceiving complex real-world environments and performing human-like tasks using on-device AI processors developed in Korea. LG Electronics will oversee the overall project and develop the humanoid system, while Mobilint will provide the NPU. Nota AI will develop the optimization software required to run AI models efficiently on the NPU within the robot.
As the lead organization for the humanoid AI model optimization workstream, Nota AI will focus on compressing and optimizing vision-language-action (VLA) models for Mobilint’s NPU, enabling fast and efficient inference directly on the humanoid robot.
Physical AI systems must continuously process visual information, understand their surroundings, make decisions and translate those decisions into actions in real time. Even small delays in VLA inference can prevent a robot from responding accurately to changing conditions. This makes optimization essential for maintaining model performance while increasing inference speed within the limited compute and power budgets of robotic systems.
Nota AI plans to apply a range of optimization techniques, including token compression, quantization, pruning, knowledge distillation and layer-wise optimization, to reduce model size, computational requirements and memory usage while minimizing performance degradation. The optimized models will then be adapted to Mobilint’s NPU architecture, taking into account the balance among accuracy, compute requirements and communication latency.
The company will also build a real-time inference environment and workload integration modules, followed by performance validation on a fully integrated system combining Mobilint’s NPU and the humanoid robot.
Running VLA models directly on the robot will allow the humanoid to perceive its surroundings, make decisions and act with less reliance on external cloud servers. Reducing data transfers between the robot and remote infrastructure can lower communication latency and improve real-time control. It can also strengthen data security by limiting the transmission of sensitive information, while optimized computation can help reduce overall power consumption.
“This project represents an important opportunity to apply Nota AI’s optimization technology to humanoid robots powered by Korean-developed NPUs,” said Myungsu Chae, CEO of Nota AI. “Building on our experience optimizing AI models across mobile, edge and data center environments, we will expand our technology and business opportunities across physical AI applications that require real-time, on-device intelligence, including humanoid robots, autonomous vehicles and industrial equipment.”
.png)