Why knowing your own position is easy, knowing the world is hard, and what closes the gap.
A robot always knows its own joints precisely — encoders report exactly where each motor is. Knowing where objects are is the hard part. Dead reckoning — assuming everything sits at fixed, pre-taught coordinates — works for a rigid, unchanging layout but drifts the moment reality differs. Vision and force sensing let a robot find things where they actually are, at the cost of more calibration and compute.
Every joint has an encoder that reports its exact angle or position, so a robot always knows the precise pose of its own arm. That part is essentially solved — the machine has perfect knowledge of itself.
The world is the problem. The simplest approach, dead reckoning, assumes every plate, tip, and tube is exactly where it was taught to be, and the robot moves to fixed coordinates. It's fast and needs no perception — but small errors accumulate, and the instant a plate is a millimetre off or a tube is in the wrong slot, the robot acts on a world that no longer matches its map [1].
Cameras let a robot see where labware actually is and re-align to it; force sensing tells it when it has made contact or met resistance. Together they let a system work on a real, variable bench — the trade is more calibration and more compute than blind coordinate-following.
This is the line between a robot that needs a perfect, fixed deck and one that copes with a working lab. A dead-reckoning system demands the world be exactly right; a vision- and force-guided system adapts to the world as it is — which is exactly why perception matters so much at the bench. Why vision matters for lab automation →
