the mistakes a robot makes can be as useful as the things it gets right
it misses the object, closes too early, comes in at a bad angle, i did all three on my first axis runs
each miss showed the same thing, the action was almost right but the movement wasn’t
that is one of the gaps in physical ai, knowing what “pick up the cup” looks like isn’t enough, the robot also needs to learn the path, timing and recovery that make the action work
@axisrobotics records that part, you control a robot arm through the browser and the whole session is saved as trajectory data rather than just a pass or fail
a clean success can be kept, a useful miss, especially when corrected, can also show where the movement went wrong and give the next attempt something to learn from
as verified runs add up, those attempts can be built toward reusable skills instead of disappearing as one off clips
a mistake stops being wasted motion, it becomes information
and that information can make the next attempt a little better
because the last attempt left a trace
you can leave one too
⟹
s.kaito.ai/l3acEja