After Teardown of Unitree G1, Japanese Engineers' Silence is More Alarming Than Concession

Last Sunday's Nikkei xTECH teardown video, I understood it as soon as I saw the third lens—when Japanese technicians used calipers to measure the thickness of the G1 joint modules, they triple-checked the readings. A colleague next to them asked, "Is this using a planetary reducer or harmonic?" He shook his head and said nothing. This silence is more convincing than any "concession" statement.

As someone who spends all day on teardown benches and production line data, what I'm focused on is a fundamental shift reflected by the G1: Humanoid robots have moved from the stage of "can we build it" to "how do we build it cheaply," and China has already established visible barriers in the latter.

Teardown of the G1: The "Brutal Aesthetics" of Cost Control

Unitree took a completely different path from Tesla Optimus and Boston Dynamics. Several key design trade-offs are clearly visible in the teardown video:

• Unified Joint Modules: Of the G1's 12 degrees of freedom, 8 use identical joint modules. This means motors, reducers, and encoders in the BOM (Bill of Materials) can be procured in bulk, making it not a dream to push the cost per set under 300 RMB. In comparison, Optimus Gen 2 has at least 5 different joint specifications, with each harmonic reducer costing over 1500 RMB.
• No Elastomer Sensing: The G1 does not use expensive six-axis force sensors but replaces force control with current loop + end-effector position estimation. This sacrifices some precision (force control resolution is around 0.5N level, while Optimus's six-axis force sensor reaches 0.05N), but it is sufficient for home and light industrial scenarios.
• Motors from the Drone Supply Chain: Unitree previously made robot dogs and drones. The rotor structure and brushless motor drive scheme of the G1 clearly show high overlap with drone motor production lines. A set of motor + driver board costs less than 200 RMB, whereas traditional robot-specific servo motors cost over 800 RMB.

Parameter Comparison	Unitree G1	Tesla Optimus Gen 2
Number of Joint Module Specs	3 types	5+ types
Cost per Joint (Est.)	280 RMB	1200 RMB
Sensor Type	Current Method + IMU	Six-Axis Force + Tactile + IMU
Reducer Type	Planetary Reducer	Harmonic Reducer
Total BOM Cost (Est.)	8000-12000 RMB	30000-50000 RMB

Looking at this table, the gap is an order of magnitude. The G1's price of 99,000 RMB (estimated launch price) is close to that of an industrial collaborative robot, but it is a bipedal humanoid. This is why the Japanese engineers were silent—they aren't unable to build it; they can't build it at this price.

Industrial Chain Foundation: This Is What Truly Made Japanese Media Concede

In the video, Japanese media reporters marveled at "China's rapid robot R&D speed," but a more accurate statement would be "China's robot supply chain response speed is fast."

There is a detail in the teardown video: The G1's joint motors use the classic 12-slot 10-pole drone motor structure with carbon fiber shells. For such motors in Shenzhen Huaqiangbei's drone parts market, the inventory turnover cycle is 7 days. If Unitree says today they want a revision, suppliers can deliver new prototypes tomorrow. In Japan, customizing such motors often requires a 4-6 week lead time.

Regarding reducers, the G1 uses planetary reducers instead of harmonic ones. Planetary reducer technology has a lower barrier to entry, with over 200 domestic suppliers, and costs less than 1/5 of harmonic reducers. Although backlash is larger than harmonic (approx. 6 arc-minutes vs. within 1 arc-minute for harmonic), this precision is entirely sufficient for humanoid robot gait control—as long as the gait algorithm is good, 6 arc-minutes of backlash can be eliminated via feedforward compensation.

For core chips, the G1's main controller is Horizon Journey 5, with 128 TOPS compute power and a cost of about 400 RMB. Optimus uses Tesla's self-developed FSD chip, with an estimated manufacturing cost of over $500 per unit. The gap lies again in the industrial ecosystem—AI chips from Horizon, Rockchip, and Allwinner have achieved large-scale shipments, and humanoid robots are just a new application scenario for these chips.

# A simple supply chain cost model
# Assuming annual production of 10,000 G1 units
BOM_cost = 10000 * 10000 # 100 million RMB
# Amortize mold fees, testing fees, labor
total_cost = BOM_cost * 1.3 # 30% overhead
# Selling price 99k, gross margin approx 40%
unit_price = 99000
revenue = 10000 * 99000 # 990 million
gross_margin = (revenue - total_cost) / revenue
print(f"Gross Margin: {gross_margin:.0%}") # Output: Gross Margin: 44%

This gross margin is normal for the consumer electronics industry but is considered exorbitant profit in the humanoid robot industry—because traditional humanoid robot companies often have negative gross margins.

Essence of Japanese Media Conceding: "Dimensional Strike" in Technical Path

Japanese media saying "we can't catch up with China" needs to be understood in context. Japan still leads in precision manufacturing (e.g., Sumitomo's harmonic reducers, NSK bearings), but the humanoid robot track has entered a competitive phase of system-level cost optimization.

The route taken by Japanese companies (like SoftBank's Pepper, Honda's ASIMO) is "top-down": pursue extreme performance first, then consider cost reduction. The result was ASIMO costing over $2 million and Pepper being discontinued. Chinese companies (Unitree, Fourier Intelligence, Agibot) take a bottom-up approach: using mature components from consumer electronics and drone supply chains to first build a machine that can run and jump with controllable costs, then improve performance through OTA and algorithm iteration.

This route is particularly intuitive in the G1: Its hardware solution is almost "FMCG" (Fast-Moving Consumer Goods) grade, but software algorithms are the true moat. Unitree uses Reinforcement Learning based on Model Predictive Control (MPC) for gait control, achieving walking stability close to Optimus on low-cost hardware. This "hardware substitution, software reinforcement" mindset is what pressures Japanese engineers the most—their strength in precision manufacturing is bypassed by Chinese companies' system integration capabilities.

Original Link: https://www.ithome.com/0/981/619.htm