Autonomy

The degree to which a robotic system makes its own decisions without human input. Autonomy is a separate dimension from the three-part robot definition. A fully teleoperated robot (where a human directs all decisions) is still a robot. A fully autonomous robot (where the robot makes all decisions) is also a robot. Autonomy level is a design parameter within a robotic architecture, not a prerequisite for the category.

The primary framework's term (p.8) for "the degree to which an agent can decide how to act towards a goal... determined by its instructions and level of human involvement." The second of the two axes defining agent capability, and the direct driver of how wide the accountability gap grows.

The degree to which an AI system operates with some independence from human involvement or control. Some level of autonomy is required under Element 2, but the definition does not require full or complete autonomy. (Recital 12; Art. 3(1))

The capacity to act toward a goal without per-step human direction. In this topic, decomposed into four distinct properties (goal-direction, planning, tool use, acting without per-step approval) rather than treated as a single switch.

The other half of the MGF's two-axis capability grid, describing how much of a task's step-by-step decision-making the agent handles versus a human. Full treatment at (see Topic 2.2).