Industry Notes / Robots: Tesla and Unitree / Chapter 5 of 10 · Thesis cards
Hardest steps: four layers
Brain unsolved industry-wide; hands and yield next
Hardest: the brain
OOD cliffs; scaling laws flat; robot data cannot be scraped.
Industry-wide unsolved—still a cost center.
Second: hand upstream
Hollow-cup motors, tendons, tactile; near half the machine’s engineering.
Thick profit while scarce; depreciates once solved.
Third: production yield
Prototype to 100k sets/year—yield engineering gates Tesla-class POs.
OEMs often push capex risk onto suppliers.
Easier than hyped
Frameless torque motors and harmonic reducers already localized.
Leaderx gross margins already slide—scarcity window closing.
Rank by tech ceiling, yield, scarcity, and qualification time. Hardest ≠ richest today.
Source：调研正文第 6 章四层难度模型；Physical Intelligence 等公开 scaling 数据
Research notes, not investment advice.