Choose a production problem
Manufacturing robots already handle material movement, assembly, welding, machine loading and unloading, painting, inspection, testing, packaging, and labeling. OSHA’s list is useful because it describes tasks, not a promise of autonomy. The practical question is whether a particular operation has repeatable inputs, observable success criteria, and enough exposure, variation, or physical burden to justify redesigning the workcell.
A robot arm is only one component. NIST describes a typical workcell as robots and controllers combined with end effectors, sensors, and supporting automation for a specific task. Fixtures, feeders, machine interfaces, inspection, guarding, and the process itself often determine whether the cell works. Treating the arm as the product hides most of the engineering and operational risk.
What is demonstrated
Bounded, repetitive operations are the strongest ground. Material handling, palletizing, welding, painting, and machine tending can constrain part presentation, paths, tooling, and access. Inspection can also be automated when cameras or gauges can see a defined feature and the pass/fail rule is measurable. These are application categories, not universal performance claims: every part family and process still needs its own evidence.
Research testbeds show what engineers can measure before deployment. NIST’s manufacturing robotics facilities study human-robot collaboration, rapid retasking, safety standards, industrial vehicles, sensors, and dexterous manipulation. NIST also publishes task boards for alignment, insertion, flexible-part handling, belt threading, and bolt threading. These are controlled measurement tools; success there is evidence about a capability under test, not proof of production uptime or quality.
Where flexibility becomes hype
General-purpose assembly remains much harder than a fixed cycle. NIST reports that many systems depend on custom hardware and software in simple, constrained environments, while people handle flexible materials, subtle alignment, and difficult fastening. Adding product variants also changes grasp points, clearances, force limits, inspection rules, and failure recovery. A polished demonstration on selected parts does not show that a cell can absorb normal production variation.
Collaboration needs the same caution. Working near a person is not one capability: coordination, communication, role allocation, human tracking, task awareness, and recovery all matter. NIST’s collaborative robotics program develops methods and metrics for those questions and explicitly notes limits in flexible factories. Calling a device ‘collaborative’ does not demonstrate that a particular application is safe or productive.
What building one involves
Define the task as states, tolerances, cycle steps, and failure modes. Select sensing and tooling around the real materials, then connect the robot to machines and production data through explicit interfaces. Measure quality, cycle time, recovery time, and changeover behavior across the expected part range. Include operators and maintenance staff in the design: OSHA says many accidents occur during programming, maintenance, testing, setup, or adjustment, when normal safeguards may be bypassed or someone enters the working envelope.
Where Vesk fits
Simulation evidence for a bounded ROS 2 task.
Vesk can help generate and test ROS 2 Jazzy code in a controlled simulation before hardware work begins. That verification is simulation-only: it does not establish physical safety, process capability, tool or sensor performance, hardware reliability, or readiness for a factory workcell.