The question
Near-body precision work such as soldering takes both hands and a fixed stand for the workpiece. A robotic third arm could hold and reorient the piece instead, but most ways of commanding one (muscle signals, foot pedals, mouth controls) add mental load. Can a robot infer what the wearer wants from where the tool is going, so nobody has to command it at all?
What was built
A camera tracks the hand-held tool and a marked workpiece; from that, the system predicts the pose the user is working towards and drives a six-axis arm to present the piece at the right angle. Twelve participants did a soldering-equivalent task under three conditions: unaided, with the arm holding the piece still like a stand, and with the arm actively repositioning it.
What was found
Holding the piece did most of the good: mental workload dropped by about 16 points on the NASA-TLX scale compared with working unaided, and usability scored above the usual benchmark. Active repositioning, however, made every participant slower (about 13 s per trial) and was not preferred over the passive stand. A clear, well-measured negative result: for this kind of task, a stable grip matters more than a clever one, and the case for active assistance has to be made with harder, more varied tasks.