[Edaily Reporter Shin Yeong-bin ] Upon disassembling the Chinese company Unitree’s humanoid robot “G1,” which retails for approximately 18 million won, the estimated cost of its internal components was found to be about 9 million won. A significant number of components—including cameras, LiDAR sensors, and the main control chip—were off-the-shelf products available on the market. Analysis suggests that the software and algorithms controlling dozens of joints to enable human-like movement are the robot’s core competitive advantage.

According to the information technology (IT) industry on the 29th, China’s Zhongyou Securities estimated the bill of materials (BOM) cost for the base model of the G1 at 41,574 yuan (approximately 8.89 million won) in a recent report. This is roughly half of the retail price of 85,000 yuan (approximately 18.17 million won), including taxes. Zhongyou Securities added approximately 3,000 yuan (about 640,000 won) in assembly and processing costs to this figure, calculating the gross profit margin for the base model at 40.7%.

“Joints Account for 66% of Cost… Price Kept Low with Inexpensive Gearboxes”

The most notable aspect of

the

G1’s

cost structure

is its joints. The report estimated the total cost of the nine large joints and 14 small joints at 27,500 yuan (approximately 5.98 million won). This accounts for approximately 66% of the total BOM cost. In effect, the joint modules—which combine motors, gearboxes, encoders, and drive boards—determine the majority of the robot’s cost.

Unitree employed relatively low-cost two-stage planetary gear reducers for its joints. While harmonic gear reducers—known for their high precision and durability—are widely used in the humanoid robot industry, Unitree appears to have secured operational performance while reducing component costs by combining its proprietary motors, drive boards, and motion control technology.

Jungwoo Securities assessed that a significant portion of the G1’s core hardware consists of mature components already in mass production. The depth camera uses Intel’s “RealSense D435i,” the LiDAR uses DJI’s “Revox Mid-360,” and the main control chip uses Rockchip’s “RK3588.” Memory and storage devices are also sourced from Chinese component manufacturers. Judging solely by the components, these are not parts that competitors would find difficult to obtain or that Unitree has exclusive access to.

On the other hand, Unitree develops its own motors, drive boards, and motion control algorithms. The report identified Unitree’s proprietary motion control algorithm as the key to the G1’s ability to walk and run quickly, as well as to regain balance after receiving an impact. While the barriers to entry for the hardware itself are not high, the software—which interprets information from each sensor in real time and controls over 20 joints as if they were a single body—is difficult to replicate in the short term.

A humanoid robot’s performance is not determined simply by the sum of its component specifications. When ground conditions change during walking or when the robot is subjected to external impacts, the torque and position of each joint must be recalculated instantly. It is also necessary to calibrate control policies based on motion data repeatedly collected from actual robots. This is why Unitree was able to perform high-difficulty movements even while using relatively inexpensive, off-the-shelf components.

9 Trillion Won in Enterprise Value Attributed to Algorithms Rather Than the “Body”

The market places a high value on Unitree’s motion control technology and system integration capabilities. This month, Unitree received approval from Chinese securities authorities for its initial public offering (IPO) registration on the STAR Market of the Shanghai Stock Exchange. The company plans to issue at least 40.45 million shares to raise approximately 4.2 billion yuan (about 900 billion won). Based on the offering size and the minimum proportion of new shares, the estimated enterprise value being discussed in the market is approximately 40–42 billion yuan, or about 8.55–8.98 trillion Hanwha. The offering price and listing date have not yet been finalized.

Unitree plans to allocate the proceeds from the offering not only to the development of robot bodies and new products, as well as the construction of production facilities, but also to research on large-scale intelligent robot models. The company aims to build upon its current motion control technology to enhance the robots’ ability to perceive their surroundings and autonomously learn new tasks. This is why the market views Unitree not merely as a robot assembler, but as a physical AI company that possesses hardware, control software, and training data.

The domestic humanoid robot industry is also accelerating efforts to secure learning and control software. LGELECTRONICS(066570)has applied a “detect-judge-act” structure to its household robot “Cloid,” which performs household tasks by detecting its surroundings, interpreting data, and then taking action. The company is enhancing the robot’s capabilities by combining information from home appliances—such as refrigerators and washing machines—with data from the robot’s sensors to enable actions tailored to the home environment.

SamsungElectronics(005930) Rainbow Robotics(277810)is also pursuing the development of intelligent humanoids by combining its bipedal walking and full-body control technologies with AI and software capabilities. The goal is to advance perception, learning, and control technologies so that robots can recognize their surroundings, determine task sequences, and move autonomously, thereby developing a robot platform suitable for real-world applications in manufacturing, logistics, and other fields.

ROBOTIS is developing the humanoid “AI Sapiens” as an open platform for imitation learning and reinforcement learning. This approach involves applying control policies learned through simulation to actual hardware and making relevant materials publicly available so that researchers can utilize the training data and software.

As humanoid robot components become more standardized and prices fall, it is becoming increasingly difficult to differentiate products based on hardware alone. Even when using the same components, the key lies in how stably a robot can walk and how quickly it can learn new tasks. Ultimately, the competitiveness of humanoid robots is expected to hinge on the ability to manufacture the body cost-effectively while making it move more precisely and intelligently.