Skip to content
←All work

Robotics & Automation

A robot arm whose encoder lied under load

A robotics startup building lab automation

eliminated

Count loss under load

4 days

Root cause found in

1

Board revisions

The problem

A pick-and-place arm held position perfectly when stationary but accumulated error during motion. The team had spent weeks in the control loop looking for a software cause. The problem was electrical: drive switching noise was corrupting encoder counts.

What we did

  1. 01

    Instrumented the encoder lines during motion to demonstrate that counts were being lost, not mis-integrated — which moved the investigation out of firmware.

  2. 02

    Separated the drive stage and the encoder interface onto distinct board regions with independent returns.

  3. 03

    Moved the encoder to a differential interface so common-mode switching noise stopped appearing as signal.

  4. 04

    Added a test that runs the arm through a worst-case motion profile and checks for count loss on every build.

Outcome

Positional accuracy under motion matched accuracy at rest. The control team recovered weeks of effort that had been aimed at the wrong layer of the system.

A worked example: a representative problem of this type, and the reasoning we would apply to it.

Next project

A charger that throttled before it reached rated power

→