Insights
Notes from the engineering side.
How we think about exceptions, confidence and running AI inside the plant. No hype, some detail.
-
· 3 min read
What an audit log for robot decisions should contain
A log that says what the robot did is not enough. The fields a decision log needs so that quality, safety and operations can answer their questions months later.
-
· 3 min read
On premise AI for robotics: what stays on site and why
Images of your products, your decisions and your logs do not need to leave the plant. What an on premise decision layer looks like in practice, and what to ask any vendor.
-
· 2 min read
Asking an operator well
When the robot is not sure, a person decides. What the tablet should show, how long the answer takes, and how to keep the question from becoming noise.
-
· 3 min read
What a 90 percent confidence should mean
A confidence score is only useful if it is calibrated. How we define calibration, how we check it against the decision log, and how it turns into a threshold.
-
· 3 min read
From a sentence to a rule: how plain language rules work
An operator types what should change. What happens between that sentence and a rule the robot follows, and why a person confirms every step.
-
· 3 min read
Why robots stop at the first exception
A robot program lists the cases it expects. Everything else is an exception. What a decision layer changes, and what it leaves alone.