Levels of Autonomy

A grading of how much a robot can perceive, decide, and act without relying on a person.

Levels of autonomy describe how much a system can handle perception, decision-making, and execution without depending on a human. The best known example is SAE International's J3016 autonomous-driving scale (2014), L0 through L5, with the key boundary between L2 and L3: from L3 up, the system, not the driver, is responsible for monitoring the environment. In robotics, Jenay Beer, Arthur Fisk, and Wendy Rogers proposed a 10-level framework in 2014 running from pure teleoperation to full autonomy, and discussed how the degree of autonomy affects how willingly people accept a robot, their situational awareness, and how they judge its reliability. In embodied AI, the common question “was this demonstration teleoperated or fully autonomous” is really asking about the level of autonomy; China's humanoid robot intelligence grading standard also borrows this leveling approach from autonomous driving. The higher the degree of autonomy, the higher the demands on robustness, failure recovery, and safety.

ExampleTwo videos might both show clothes being folded, but one where an operator wears a VR headset and controls the robot in real time sits at the lowest level of autonomy; only a robot that sees, decides, and acts entirely on its own, with no human involved, counts as fully autonomous.

Also called

Autonomy

Related

Fully Autonomous · Teleoperation · Humanoid Robot Intelligence Level Grading · Shared Autonomy · Human-Robot Interaction · Autonomous Driving

Sources

Toward a framework for levels of robot autonomy in human-robot interaction (Beer, Fisk, Rogers, 2014)
Self-driving car - Wikipedia（SAE J3016 分级） (Chinese)