Putting a small network connector into its socket is harder than it looks. The plug must face the right way, line up and enter without slipping. A team led by Ammar Abbas compared ways to teach a five-fingered robot hand this task. They carried out roughly 1,700 attempts on real equipment, not just on a screen.
The team tested 281 versions of control software from seven groups. They changed what information the program received from cameras and how it described finger movements. This is called a data representation[1]: the same task can be described using different pictures and numbers. Sometimes that choice mattered more than ordinary adjustments to the training process.
The researchers also examined 67 scores that could be calculated without performing the insertion. Only five, related to how smooth and direct the predicted motion was, closely matched real success. Even those worked only within one group of programs. A movement that looks excellent in a computer test may still miss the actual socket.
For point-cloud policies, cropping the relevant region, using robot-base coordinates and normalizing features improved success by 70 to 100 percentage points in individual controlled comparisons. For a vision-language-action policy, absolute rather than relative actions improved insertion success by up to 70 points. These figures come from different configurations and cannot be added together or generalized to every architecture. Seven policy families were included, exposing how representation choices interact with model design.
The findings concern particular hands, connectors and test conditions. They do not establish the best program for every object. They do explain why a robot must be tested with the things it will really handle.
For a customer automating assembly, this points to acceptance tests on actual components. Software may score well internally yet still miss a socket. A procurement specification should define connector geometry, tolerances, trial counts, cycle time and lighting. The paper offers no universal deployment price or guarantee for other connectors, and its most striking improvements arise from particular representation choices rather than a new robot purchase.
Only five of 67 offline measures reliably correlated with real insertion success, and only within one policy family. Correlations between minus 0.70 and minus 0.80 for predicted motion smoothness and directness do not turn these scores into substitutes for hardware trials. Selecting a controller from validation loss alone is particularly weakly supported here.
Only five of 67 offline measures predicted success, and only within one policy family. Replacing physical acceptance testing with a single model score is therefore not supported by these findings.