Mandy Meindersma posted a project to Hackster.io that begins with a deceptively simple question: can a uFactory robot arm juggle?
The arm is known for precision. Juggling is not a precision problem in the sense that industrial robotics means it, and that gap is the whole project. Meindersma worked with Jie Xuan Tang and Aika K at Viam’s Fine Motor Skills Hackathon in September 2026 to find out.
Their own summary of the outcome: much harder than they thought.
Why an accurate arm is bad at this
An industrial arm’s design goal is to reach a commanded pose and hold it, repeatably, against gravity and payload. Everything about it — the gearing, the control loop, the trajectory planner — is tuned for arriving accurately.
Juggling asks for something close to the opposite. The useful moment is not an arrival but a release, and what matters is the object’s state at the instant contact ends: its velocity vector, its spin, the exact millisecond the grip opens. After that the arm has no authority over it at all. The ball is on a ballistic arc, and the arm’s only remaining job is to be somewhere else, early, with its hand open.
That reframes the problem from positioning to timing and prediction, which is where the hard parts live:
- A trajectory planner that optimises for smooth arrival at a coordinate is not optimising for a specific velocity through a point.
- Gripper actuation latency becomes a first-order term. A few milliseconds of variation in when the fingers open is a large change in where the ball lands.
- Catching requires estimating the arc from partial observations and committing to an intercept before the object gets there — the arm cannot react at the last moment, because it has mass.
None of this is what the hardware was built for, which is precisely why it is an interesting thing to attempt in 72 hours.
Failure as the deliverable
What makes the write-up worth reading is that it is honest about the result. The team has published their code for anyone who wants to see how they drove the arm, and the repository is the artefact — not a polished demo reel.
This is an underrated genre in hardware documentation. A project that works tells you one path through the problem. A project that does not work, documented properly, tells you where the walls are, and that is usually the more transferable information. Anyone else who points an industrial arm at a ballistic task will hit the same gripper-latency and trajectory-planning questions, and will now hit them with a head start.
The hackathon format
Viam’s Fine Motor Skills Hackathon put software engineers in front of robot arms with a short clock, which produces a particular kind of project: ambitious in concept, constrained in execution, and unusually candid about what went wrong because there was no time to hide it.
The genre has a long creative-technology pedigree. Making a machine do something it was never specified for — juggle, draw, play an instrument, throw — is how a surprising amount of what we now consider normal robot capability got started, and the arms that juggle badly this year are the ones whose control problems get solved next.