Robot anatomy: body, sensors, actuators and compute
Post388 words
Reduces any robot to four blocks in a closed loop: mechanics set hard limits, sensors set the autonomy ceiling, actuators determine torque and bandwidth, and compute runs control loops far beyond 'AI'.
๐ฅ๐ผ๐ฏ๐ผ๐ ๐ฎ๐ป๐ฎ๐๐ผ๐บ๐: ๐ฏ๐ผ๐ฑ๐, ๐๐ฒ๐ป๐๐ผ๐ฟ๐, ๐ฎ๐ฐ๐๐๐ฎ๐๐ผ๐ฟ๐, ๐ฐ๐ผ๐บ๐ฝ๐๐๐ฒ
A robot is easiest to understand as a closed loop: it measures the world, decides what to do, and applies forces back into the world. If you strip away form factors (arm, drone, wheeled base, humanoid), almost every robot reduces to four blocks:
โ ๐๐ผ๐ฑ๐ (๐บ๐ฒ๐ฐ๐ต๐ฎ๐ป๐ถ๐ฐ๐) - ๐๐ต๐ฎ๐ ๐๐ต๐ฒ ๐ฟ๐ผ๐ฏ๐ผ๐ ๐ฐ๐ฎ๐ป ๐ฝ๐ต๐๐๐ถ๐ฐ๐ฎ๐น๐น๐ ๐ฑ๐ผ
The body is the structure: links/joints, wheels/legs, frames, transmissions, and end-effectors. Mechanics sets the hard limits:
โค payload (how much force/weight it can handle)
โค reach/workspace (where it can act)
โค precision (how repeatable motion is)
โค speed + stability (how fast without oscillation or tipping)
โค contact capability (can it push, pull, grasp, or only โmove aroundโ?)
If the mechanics arenโt there (stiffness, torque, traction), no amount of software will โfixโ it.
โ ๐ฆ๐ฒ๐ป๐๐ผ๐ฟ๐ - ๐๐ต๐ฎ๐ ๐๐ต๐ฒ ๐ฟ๐ผ๐ฏ๐ผ๐ ๐ธ๐ป๐ผ๐๐
Sensors turn reality into signals. They usually split into:
โค proprioception (internal state): encoders, motor current, torque sensors, IMU - โwhat is my body doing?โ
โค exteroception (external world): cameras, LiDAR, radar, depth - โwhatโs around me?โ
Sensors define your ceiling for autonomy. If your perception is weak (bad lighting, occlusions, noisy IMU), your robot will either move slowly and cautiously or fail often.
โ ๐๐ฐ๐๐๐ฎ๐๐ผ๐ฟ๐ - ๐๐ต๐ฎ๐ ๐๐ต๐ฒ ๐ฟ๐ผ๐ฏ๐ผ๐ ๐ฐ๐ฎ๐ป ๐ฎ๐ฝ๐ฝ๐น๐ ๐๐ผ ๐ฟ๐ฒ๐ฎ๐น๐ถ๐๐
Actuators are the muscle: motors, hydraulics, pneumatics, sometimes artificial muscles. They convert electrical/fluid power into motion/force.
Actuators largely determine:
โค torque/force (can it lift/open/hold?)
โค bandwidth (how fast it can respond)
โค backdrivability / compliance (how safe and controllable it is around people)
โค efficiency + heat (how long it can sustain work before thermal limits)
โ ๐๐ผ๐บ๐ฝ๐๐๐ฒ + ๐ฐ๐ผ๐ป๐๐ฟ๐ผ๐น - ๐ต๐ผ๐ ๐๐ต๐ฒ ๐ฟ๐ผ๐ฏ๐ผ๐ ๐๐๐ฟ๐ป๐ ๐๐ถ๐ด๐ป๐ฎ๐น๐ ๐ถ๐ป๐๐ผ ๐ฎ๐ฐ๐๐ถ๐ผ๐ป
Compute is the โbrainโ that runs estimation, planning, and control. In real robots, a big part of compute is not โAIโ - itโs control loops and hardware interfaces that read sensors and write actuator commands at fixed rates.
ROS 2 control frameworks describe this as a control loop that repeatedly calls read() (get hardware state) and write() (send commands) through a hardware abstraction layer.
Compute quality maps to capabilities like:
โค latency (can it react fast enough to stay stable?)
โค robustness (does it fail gracefully on sensor dropouts?)
โค autonomy depth (simple teleop - assisted autonomy - fully autonomous behaviors)
โค fleet operations (logging, updates, monitoring)
1 image on X
What came next ยท since 08 Jan 2026
Since then
Zero has written 1 more piece on ROS 2 since this one.