What does autonomy mean in practice?
A measurement guide to autonomy that makes room for supervision, teleoperation, task scope and interventions.
Maintained reference ยท Updated 2026-10-20
Autonomy is a spectrum of responsibility
A robot may perform a movement locally while a person chooses the task, approves each step, handles exceptions or remotely takes control. Calling a system simply 'autonomous' hides those distinctions. Describe the allocation of work between robot, operator and surrounding process instead.
Define the task boundary
A meaningful result says what starts and ends the task, what counts as success, and what the robot was allowed to do. 'Picked up an object' is narrower than 'completed an order'; 'worked in a prepared cell' is not equivalent to navigating an unmodified facility.
Count interventions and attempts
A completion rate needs a denominator: how many attempts, under what conditions, and how were retries counted? Report interventions separately where possible, including pauses, remote assistance and resets. A single successful run demonstrates possibility, not reliability.
Account for the operating envelope
Lighting, floor surfaces, object variation, people nearby, network access and task duration all constrain performance. Specify which conditions were tested and which were not. An operating envelope is more useful than an unqualified claim of generality.
Compare like with like
When evaluating results, align protocol, task version, sample size, hardware, software and reporting source. Vendor-reported results and independently conducted tests can both be informative, but they are not interchangeable. The source and method should travel with the number.