RobotGuide

Selection

Hardware selection

Component selection is not about picking the highest specs; it is about picking the set whose weakest link is not fatal. This section does one thing: clarify what to decide first and what to decide later, then give tables you can compare against. Prices are common ranges from public quotes on domestic e-commerce and distributors in 2026, and they move with volume and exchange rates.

The five tables are ordered: fix the controller first, then actuation and sensing, then instruments. Wrong order, money spent twice.

MCU and Domain Controller

How to actually choose between STM32, Raspberry Pi, and Jetson

6 rows · 1 / 5

Motor and Transmission

First compute the end-effector torque, then work back to the motor

6 rows · 2 / 5

Sensor

What to measure, how fast, how accurate

9 rows · 3 / 5

Boards, Instruments, and Tools

Buy what lets you measure first, then what lets you build

9 rows · 4 / 5

Three Budget Tiers

What 3,000 / 15,000 / 50,000 RMB can each achieve

3 rows · 5 / 5

Four universal criteria

CriterionHow to use itCost of getting it wrong
Control frequencyAsk first: «how many Hz in the innermost loop». Above 1 kHz, MCU territory; only consider Linux at tens of Hz.Paid more and still cannot do it; rework onto another platform
Interfaces and protocolsWhich bus the joints use, what interface the camera needs, whether sync signals are required.Interfaces run out; adapters add failure points
Power and thermalsMobile platforms: reserve peak power × 1.5; continuous torque is half of peak.Thermal throttling mid-run looks like «the algorithm got worse»
TestabilityTest points provision? Waveforms scope-able? Logs sufficient?Built but not debuggable; all time spent guessing

One honest sentence

Selection tables do not hand you the answer; they show you which item is negotiable. Real projects never have every item optimal — being able to state «I compromised this one, and here is why that is acceptable» is worth more than owning the best-specced device.