Robotics fundamentals

In one line

Every robot runs the same loop: sense → plan → act. Understand those three words and the rest is engineering detail — including why a robot that welds car doors flawlessly cannot be trusted near a person.

The loop

Sense — where am I, where is everything else? Cameras, encoders, force sensors, LIDAR. The robot's world model is only as good as this, and sensors lie in specific ways.

Plan — given where I am and where I want to be, what's the path? This is where the maths lives: kinematics (where does the tip go if I move these joints?), inverse kinematics (what joint angles put the tip there?), and path planning around obstacles.

Act — actuators move, control systems keep the movement true, and the sensors report the result. Loop again, often a thousand times a second — which is why these systems run an RTOS where "usually fast enough" isn't a category.

Degrees of freedom, and the wrist

Degrees of freedom (DOF) is how many independent ways a thing can move. Your shoulder has 3; a typical industrial arm has 6, which is the minimum to reach any position at any orientation in space.

This is where surgical robots get interesting. A laparoscopic instrument through a fixed port loses DOF — it can go in, out, rotate, and pivot, but the port constrains everything. The celebrated innovation of surgical robots was putting the wrist back inside the patient: articulation at the tip restores the dexterity the port took away. That's the actual advance — not autonomy, not intelligence, but recovering degrees of freedom that laparoscopy had removed.

What makes medical robotics different

Factory robotics solves a hard problem in an easy environment. Medical robotics solves a hard problem in a hostile one:

• The environment is a person. It deforms, breathes, bleeds, and moves. A factory jig is in the same place every time; an organ is not.
• The workspace is unmapped. Anatomy varies between people and differs from the imaging you planned on, because the imaging was taken when they were lying differently.
• You cannot iterate. A factory robot's failure scraps a part. There is no scrap here.
• Safety is regulatory and physical. ISO 13482 for care robots, IEC 60601 for the electrical envelope, plus the device pathway.
• Sterility. A constraint with no analogue in industry: every part near the patient must be sterilisable or disposable, and that dictates mechanical design far more than the software.

On autonomy

The demos suggest we're close. We are not, and it's worth understanding why.

Almost every surgical robot in clinical use today is teleoperated — a master–slave system. The surgeon moves; the robot mirrors it, filtering tremor and scaling motion. It is a very good instrument. The intelligence is the human's, and every decision is theirs.

The gap to autonomy isn't dexterity — machines out-manoeuvre us already. It's perception and judgement: recognising unexpected anatomy, deciding that the plan is now wrong, knowing when to stop. That is the same wall clinical AI hits, with the consequence of error made immediate and physical.

Expect the trajectory to run through shared autonomy — the robot enforcing a no-fly zone around a vessel, holding a plane steady, or executing a well-defined sub-task under supervision — long before anything closes the loop on a whole operation.

Where informatics touches it

The robot is a data source: kinematics, forces, video, timing, every instrument exchange. That stream is a rich, almost untouched record of how an operation was actually performed — which is why surgical data science exists, and why the interesting questions here are about provenance, consent and comparability rather than about motors.