Introduction
Autonomous systems do not all possess the same degree of independence. Some systems only assist a human operator, while others can perceive their environment, make decisions, and complete tasks with very limited human involvement. The degree to which a system can perform a task independently is commonly described as its level of automation or level of autonomy.
Understanding these levels is important because the word “autonomous” is often used too broadly. A vehicle with automatic braking is not necessarily a fully autonomous vehicle. Similarly, a robot that follows a fixed route is not equivalent to a robot that can map an unknown environment, identify obstacles, select a new route, and complete a task independently.
The level of autonomy depends on how responsibilities are divided between the human operator and the machine. At lower levels, the human performs most of the decision-making and control. At higher levels, the system takes responsibility for perception, planning, decision-making, and action within defined operating conditions.
Levels of autonomy help engineers, users, researchers, regulators, and organizations understand what a system can do, when human involvement is required, and what safety measures must be established.
Understanding Levels of Automation
Automation refers to the use of technology to perform specific functions that would otherwise require human effort. The level of automation describes how much of a task is performed by the machine and how much remains under human control.
A low-level automated system may perform only one function, such as maintaining a selected speed. A more advanced system may control speed, steering, and braking simultaneously. A highly autonomous system may also monitor the environment, predict risks, select a route, respond to unexpected events, and determine when human intervention is necessary.
Automation levels can therefore be understood as a gradual transfer of responsibility from the human operator to the machine.
At the lowest level, the human performs the task completely. The machine may provide information but does not control the operation. At an intermediate level, the system assists with selected functions while the human remains responsible for supervision and decision-making. At higher levels, the system manages the complete task within specific conditions. At the highest level, the system is expected to operate without human involvement across a wide variety of environments.
This progression does not mean that a higher level is always better. The appropriate level depends on the application, environment, safety requirements, cost, technical capability, and consequences of failure.
For example, a highly autonomous cleaning robot may be suitable for a controlled indoor environment. The same level of independence may not be appropriate for a medical robot performing a critical surgical procedure. In healthcare, human supervision may remain essential even when advanced automation is available.
Human and Machine Responsibility
The central difference between levels of autonomy is the distribution of responsibility between the human and the system.
Human responsibility may include observing the environment, selecting goals, making decisions, controlling movement, monitoring system performance, responding to failures, and approving critical actions.
Machine responsibility may include collecting sensor data, identifying objects, estimating location, planning movement, controlling actuators, detecting hazards, and correcting errors.
At lower levels of autonomy, the machine performs limited functions and the human remains responsible for the overall task. At higher levels, the system takes control of more functions and may be responsible for monitoring both the environment and its own performance.
The transfer of responsibility must be clearly defined. Unclear responsibility can create serious safety problems. A human may incorrectly assume that the system is controlling a situation, while the system may be expecting the human to intervene.
This problem is especially important in vehicles, aircraft, industrial robots, medical systems, and other safety-critical applications. Users must understand what the system can do, what it cannot do, and when they are expected to take control.
Manual Operation
Manual operation represents the lowest level of system autonomy. The human operator performs the complete task and directly controls the system.
A manually driven car is an example. The driver observes the road, identifies hazards, selects the route, controls acceleration and braking, and operates the steering system.
A remotely controlled drone is another example. The operator determines the direction, altitude, speed, and movement of the drone using a control device. The drone may stabilize itself automatically, but the main navigation decisions are made by the human.
In manual operation, technology may still provide information or basic support. A vehicle dashboard may display speed, fuel level, or engine warnings. These features assist the driver but do not independently control the driving task.
The human remains responsible for perception, decision-making, planning, and action.
Driver or Operator Assistance
At the assistance level, the system performs one or more limited functions while the human remains fully responsible for the overall operation.
Examples include cruise control, parking sensors, lane-departure warnings, automatic lighting, and collision alerts.
Cruise control can maintain a selected speed, but the driver must continue steering, observing traffic, and applying the brakes when required. Parking sensors warn the driver about nearby obstacles, but the driver remains responsible for moving the vehicle safely.
In industrial environments, an assistance system may provide positioning guidance to a machine operator. A robotic arm may compensate for vibration while the human controls its main movements.
Operator-assistance systems are designed to reduce workload, improve accuracy, or increase safety. However, they do not replace human observation and judgment.
The human must remain attentive because the system normally cannot manage the complete task or respond to every unexpected situation.
Partial Automation
Partial automation occurs when the system can perform several control functions at the same time, but the human must continuously monitor the environment and remain prepared to intervene.
In a partially automated vehicle, the system may control acceleration, braking, and steering under certain conditions. However, the driver remains responsible for observing traffic, identifying hazards, and taking control whenever necessary.
A drone may automatically follow a planned route while a remote operator monitors its flight. The drone may regulate altitude and direction, but the human must respond if communication is lost, weather conditions change, or an unexpected obstacle appears.
In manufacturing, a robot may automatically perform a sequence of movements while an operator supervises the work area and manages unusual conditions.
Partial automation can reduce physical and mental workload, but it may also create overreliance. When the system performs well for long periods, the human operator may become distracted or may overestimate its capabilities.
Safe partial automation therefore requires clear instructions, effective warning systems, and realistic communication about system limitations.
Conditional Automation
Conditional automation allows the system to perform the complete operational task within specific conditions. The system monitors the environment, controls the machine, and makes routine decisions. However, it may request human intervention when it reaches the limits of its capability.
For example, an automated vehicle may control driving on a mapped highway under clear weather conditions. It may monitor traffic, maintain lane position, adjust speed, and respond to nearby vehicles. If heavy rain begins or road markings become unclear, the system may ask the driver to take control.
A conditionally autonomous drone may complete an inspection mission independently but request remote assistance if it detects an unfamiliar object or experiences a sensor problem.
The main challenge at this level is the transfer of control between the machine and the human.
The system must provide sufficient warning before human intervention is required. The human must have enough time to understand the situation and take appropriate action. A sudden request for intervention may be unsafe if the operator has not been actively monitoring the environment.
Conditional automation requires careful design of alerts, interfaces, control-transfer procedures, and fallback actions.
High Automation
At the high-automation level, the system can perform the complete task and manage failures within a defined operating environment. Human intervention is not normally required during operation within that environment.
An autonomous shuttle operating inside a university campus may represent high automation. It may navigate along mapped routes, recognize pedestrians, avoid obstacles, stop at designated locations, and perform emergency braking.
If the shuttle encounters a situation it cannot manage, it should move into a safe condition. It may stop at the roadside, return to a designated area, reduce speed, or contact a remote supervisor.
A warehouse robot may also operate with high autonomy inside a controlled facility. It may receive delivery tasks, plan routes, avoid people and other robots, monitor battery status, and return to a charging station without human assistance.
High automation is normally limited to a clearly defined Operational Design Domain. The system may work reliably in a warehouse, campus, mine, port, farm, or mapped industrial environment but may not be capable of operating safely outside that domain.
The human does not need to supervise every movement, but human support may still be required for maintenance, mission planning, unusual failures, or remote assistance.
Full Automation
Full automation represents the highest theoretical level of autonomy. A fully autonomous system can perform all aspects of a task under a wide range of environmental and operational conditions without requiring human intervention.
A fully autonomous vehicle would be able to drive on highways, urban roads, rural roads, and unfamiliar routes during different weather, lighting, and traffic conditions. It would manage navigation, perception, decision-making, control, emergency situations, and system failures independently.
A fully autonomous robot would be capable of understanding new environments, selecting suitable actions, adapting to unfamiliar conditions, and completing its mission without human supervision.
Achieving full autonomy is extremely difficult because real-world environments contain unpredictable people, rare events, changing weather, incomplete information, damaged infrastructure, unusual objects, and complex ethical situations.
Full autonomy also requires strong reliability, cybersecurity, fault tolerance, self-monitoring, and safe fallback behaviour.
For this reason, many systems described as autonomous are not fully autonomous. They operate independently only within restricted conditions and still depend on humans for supervision, maintenance, or emergency support.
Levels of Driving Automation
Autonomous vehicles are commonly described through six levels, ranging from Level 0 to Level 5. These levels show how responsibility for the driving task is distributed between the driver and the automated driving system.
At Level 0, there is no sustained driving automation. The driver controls steering, acceleration, and braking. The vehicle may provide warnings or brief safety interventions, but it does not continuously perform the driving task.
Examples include collision warnings, blind-spot alerts, and lane-departure warnings. These systems inform or briefly assist the driver, while the driver remains fully responsible.
At Level 1, the vehicle provides assistance with either lateral control or longitudinal control. Lateral control generally refers to steering, while longitudinal control refers to acceleration and braking.
A vehicle may provide lane-centering assistance or adaptive cruise control, but not both as a complete combined driving function. The driver must continuously supervise the road and remain responsible for all other aspects of driving.
At Level 2, the system can control steering, acceleration, and braking at the same time under certain conditions. The driver must continue monitoring the road and must be ready to take control immediately.
This level is often described as partial driving automation. Although the system performs multiple driving functions, the driver remains responsible for environmental observation and safe operation.
At Level 3, the automated system performs the complete driving task within specific conditions and monitors the surrounding environment. The driver may temporarily stop supervising the road continuously but must be available to take control when requested.
This is known as conditional driving automation. Safe transfer of control is one of the most difficult challenges at this level because the driver may need time to understand the road situation before responding.
At Level 4, the system performs the complete driving task without expecting a human to intervene within its defined Operational Design Domain.
If the system reaches its limits, it must achieve a safe condition independently. A Level 4 autonomous shuttle may operate only on specific roads, within a mapped city zone, under selected weather conditions, or at limited speeds.
Human control may be possible, but it is not required during normal operation within the approved domain.
At Level 5, the system can perform the complete driving task under all conditions in which a human driver could reasonably operate. It is not restricted to a particular route, geographic area, road type, or environmental condition.
A Level 5 vehicle would not require a human driver, steering wheel, or driving controls. This level represents complete driving autonomy and remains a major technical challenge.
Autonomy Beyond Vehicles
Although vehicle automation levels are widely known, autonomous systems in other industries require different ways of describing autonomy.
A drone may be evaluated according to whether the human controls every movement, provides only a flight route, monitors the mission, or has no involvement after launch.
An industrial robot may be classified according to whether it repeats a fixed motion, responds to sensor information, selects tasks, coordinates with other robots, or adapts to changing production conditions.
A medical robot may be evaluated according to whether it provides information, guides a clinician, performs a selected action, or independently makes and executes a treatment decision.
A software agent may be classified according to whether it recommends actions, executes approved actions, independently selects actions, or modifies its own strategy.
Therefore, autonomy levels should always be interpreted in relation to the task being performed. A system may be highly autonomous in one function but have very low autonomy in another.
For example, a drone may fly automatically along a planned route but depend on a human to decide what data should be collected. A robot may navigate independently through a warehouse but require a human to assign each delivery task.
Autonomy is therefore multidimensional rather than a single fixed property.
Task Autonomy and System Autonomy
Task autonomy describes the extent to which a system can complete a particular function independently.
A robotic arm may have high autonomy when sorting objects but no autonomy when selecting production goals. A vehicle may automatically park but may not drive autonomously on public roads.
System autonomy refers to the overall independence of the complete system. It includes perception, decision-making, planning, control, fault handling, communication, and mission management.
A system should not be described as fully autonomous merely because one component operates automatically.
For example, automatic braking is an autonomous safety function, but it does not make the complete vehicle autonomous. Similarly, obstacle avoidance does not make a drone fully autonomous if a human still controls its route and mission.
Accurate descriptions of autonomy should identify the specific task, environment, and conditions in which the autonomous capability is available.
Human-in-the-Loop Systems
In a human-in-the-loop system, the human directly participates in the decision-making process.
The autonomous system may analyse data, identify possible actions, or generate recommendations, but the human approves or selects the final action.
A medical decision-support system may analyse a patient’s data and recommend possible diagnoses, but the clinician makes the final decision.
A military surveillance system may detect a potential object of interest, but a human operator must verify the detection and authorize further action.
Human-in-the-loop systems are useful when decisions have serious ethical, legal, financial, or safety consequences.
Human involvement can provide judgment, accountability, contextual understanding, and ethical oversight. However, it may also slow the system and introduce human error.
Human-on-the-Loop Systems
In a human-on-the-loop system, the machine performs tasks and makes routine decisions independently while a human supervises its operation.
The human does not approve every action but can intervene, stop the system, or modify its behaviour when necessary.
A fleet of warehouse robots may operate independently while one operator monitors their status from a control centre. The operator becomes involved only when a robot encounters a fault or unusual situation.
A drone may complete an inspection route automatically while a remote pilot monitors its flight and remains ready to take control.
Human-on-the-loop systems can provide higher efficiency than human-in-the-loop systems while retaining a level of human oversight.
The effectiveness of supervision depends on clear information, manageable workload, reliable alerts, and sufficient time for intervention.
Human-out-of-the-Loop Systems
In a human-out-of-the-loop system, the machine makes and executes decisions without immediate human involvement.
The human may define the mission, operating rules, and safety limits before operation, but does not participate in real-time decisions.
A planetary rover may navigate around an obstacle without waiting for instructions from Earth. An underwater robot may adjust its route when communication with the surface is unavailable.
Human-out-of-the-loop operation may be necessary when communication is delayed, unavailable, or impractical.
However, this model requires strong safety controls because the system must manage uncertainty and failures independently. It also raises important questions about accountability, ethical limits, and appropriate human control.
Shared Autonomy
Shared autonomy is an approach in which the human and machine cooperate to perform a task. Control is divided dynamically according to the situation, system capability, and user needs.
A powered wheelchair may allow the user to select a direction while the autonomous system avoids obstacles and maintains a safe path.
A surgical robot may allow the surgeon to control the main movement while the system filters hand tremors, prevents unsafe motion, and maintains tool stability.
A drone operator may select a target location while the system calculates the route, maintains altitude, and avoids obstacles.
Shared autonomy combines human judgment with machine precision. It is especially useful when complete automation is not technically possible or when human decision-making remains important.
The main challenge is determining when control should belong to the human, when it should belong to the machine, and how control should be transferred smoothly.
Supervisory Control
Supervisory control allows a human to provide high-level instructions while the system performs lower-level actions independently.
Instead of directly controlling every movement, the human defines goals such as “inspect this building,” “move this package,” or “travel to this location.”
The autonomous system then determines how to complete the task. It may select a path, control movement, avoid obstacles, and monitor its own condition.
Supervisory control is common in robotics, space exploration, mining, industrial automation, and drone operations.
It reduces the operator’s workload because the human focuses on mission-level decisions rather than continuous control.
However, the autonomous system must communicate its status clearly so that the supervisor understands what it is doing, why it selected a particular action, and whether assistance is required.
Operational Design Domain and Autonomy Level
A system’s autonomy level cannot be understood without considering its Operational Design Domain.
The Operational Design Domain defines the conditions under which the autonomous function is designed to operate. These conditions may include location, weather, lighting, speed, terrain, road type, communication availability, and traffic conditions.
A system may demonstrate high autonomy inside a controlled warehouse but have no autonomous capability outdoors.
An autonomous shuttle may operate without a driver on a mapped campus but may require manual transport when moving to another location.
A drone may perform autonomous inspection in low-wind conditions but require human control during strong winds.
The autonomy level therefore applies only within the specified operating domain.
When the system moves outside that domain, it may reduce its level of automation, request human intervention, stop safely, or disable the autonomous function.
Fallback and Minimum-Risk Conditions
Higher levels of autonomy require the system to respond safely when it can no longer continue normal operation.
A fallback response is the action taken when a failure, unexpected event, or operating limitation prevents the system from completing its task normally.
A vehicle may reduce speed and stop in a safe location. A drone may return to its launch point or perform a controlled landing. A mobile robot may stop and request assistance. An industrial machine may enter a safe shutdown state.
The system should not simply stop operating without considering the surrounding environment. An immediate stop may be safe for a factory robot but unsafe for a vehicle travelling in traffic.
A minimum-risk condition is the safest practical state the system can achieve under the circumstances.
The appropriate fallback response depends on the application, environment, system condition, and type of failure.
Transfer of Control
Control transfer occurs when responsibility moves from the human to the automated system or from the system back to the human.
A safe transfer requires clear communication. The user must understand when the system has taken control, what functions it is managing, and when the user is expected to intervene.
The system should provide visual, auditory, or physical alerts when control must be transferred.
The human also needs sufficient time to understand the situation. A person who has not been monitoring the environment may require several seconds to observe conditions, identify hazards, and make a safe decision.
Poorly designed control transfer can lead to mode confusion. Mode confusion occurs when the human misunderstands which functions are active or who is responsible for the task.
For example, a driver may believe that a vehicle is controlling steering when only speed control is active. An operator may assume that a robot will avoid all obstacles even though obstacle avoidance is available only at low speeds.
Clear interface design and user training are therefore essential.
Trust and Overreliance
The effectiveness of autonomous systems depends partly on whether users trust them appropriately.
Insufficient trust may cause users to avoid useful autonomous functions or intervene unnecessarily. Excessive trust may cause users to ignore warnings, stop monitoring the system, or use it outside its intended conditions.
Appropriate trust means that the user understands both the capabilities and limitations of the system.
Autonomous systems should communicate uncertainty, operating status, active functions, and limitations clearly. Users should not be encouraged to believe that a partially automated system is fully autonomous.
Training, transparent system behaviour, effective alerts, and accurate marketing descriptions are important for preventing misuse and overreliance.
Factors That Determine the Level of Autonomy
The achievable level of autonomy depends on several connected factors.
The complexity of the environment is important. A structured warehouse with mapped pathways is easier to automate than a crowded public street.
The predictability of the task also matters. Repetitive industrial tasks can often achieve high automation, while tasks involving human emotions, ethical judgment, or highly variable conditions may require continued human involvement.
Sensor quality influences how accurately the system understands its environment. Decision-making algorithms determine whether it can select appropriate actions. Processing capability affects whether decisions can be made in real time.
Reliable actuators, communication systems, control algorithms, cybersecurity measures, and fault-handling mechanisms are also necessary.
The consequences of failure affect how much autonomy is acceptable. A minor failure in a household cleaning robot may be inconvenient, while a failure in an autonomous aircraft or medical robot may be life-threatening.
Legal, ethical, organizational, and economic factors may also limit the level of autonomy even when the technology is technically capable.
Advantages of Higher Automation
Higher levels of automation can reduce human workload, improve consistency, increase operational efficiency, and enable systems to work in dangerous or inaccessible environments.
Autonomous machines can perform repetitive tasks for long periods without fatigue. They may respond faster than humans in certain situations and can continuously monitor large quantities of sensor data.
High autonomy can also support remote operations in mines, oceans, disaster areas, contaminated environments, and space.
In transportation and logistics, autonomous systems may optimize routes, reduce delays, and coordinate multiple vehicles or robots.
However, these benefits depend on reliable design, appropriate testing, and operation within defined conditions.
Challenges of Higher Automation
As the level of autonomy increases, the technical and safety challenges also become greater.
The system must understand increasingly complex environments, make decisions under uncertainty, predict the behaviour of people and objects, and manage failures without human assistance.
Testing becomes more difficult because the system may encounter an enormous number of possible situations.
Higher autonomy also creates questions about responsibility. If an autonomous system causes harm, responsibility may involve the developer, manufacturer, operator, owner, data provider, or organization deploying the system.
Cybersecurity becomes increasingly important because a highly autonomous system may control physical equipment and make critical decisions.
Human skills may also decline when people rely heavily on automated systems. An operator who rarely controls a machine manually may find it difficult to respond effectively during an emergency.
For this reason, higher automation should not be introduced only because it is technically possible. It should be introduced when it provides a clear benefit and can be operated safely, responsibly, and reliably.
Selecting an Appropriate Level of Autonomy
The ideal level of autonomy depends on the application rather than on the goal of achieving complete independence.
A system should use the level of autonomy that provides the best balance among performance, safety, cost, human oversight, and operational complexity.
In a controlled warehouse, high autonomy may be practical because the environment can be mapped and access can be managed.
In surgery, shared autonomy may be more appropriate because the machine can provide precision while the clinician maintains decision-making authority.
In aviation, advanced automation may control routine flight operations while trained pilots remain responsible for supervision and emergencies.
In agriculture, an autonomous tractor may operate independently in a field but require a human to define the task, inspect the equipment, and respond to unusual conditions.
The objective should therefore be appropriate autonomy rather than maximum autonomy.
Conclusion
Levels of automation and autonomy describe how responsibility for a task is shared between humans and machines.
At lower levels, the human performs the main task while the system provides information or limited assistance. At intermediate levels, the system controls several functions but requires continuous supervision. At higher levels, the system performs the complete task within defined conditions and may manage failures without immediate human intervention.
Autonomy can involve human-in-the-loop, human-on-the-loop, human-out-of-the-loop, shared-autonomy, and supervisory-control arrangements.
The level of autonomy must always be considered together with the system’s specific task and Operational Design Domain. A machine may be highly autonomous in one function or environment but dependent on human control in another.
Higher autonomy can improve efficiency, consistency, accessibility, and safety in dangerous environments. However, it also creates challenges related to perception, control transfer, cybersecurity, reliability, accountability, human trust, and failure management.
The most suitable autonomous system is not necessarily the one with the highest possible level of automation. It is the system that uses an appropriate division of responsibility between the human and machine while operating safely, reliably, and effectively within its intended conditions.
Comments are closed.