Over the past several decades, autonomous vehicles have gone from a futuristic idea in books and movies to being available and deployed on the road. Nowadays, we have vehicles that can operate with minimal supervision, such as Ford’s BlueCruise technology for hands-free highway driving, to human-less robotaxis, such as those offered by Cruise in San Francisco, Austin, and Phoenix (only available at night). Moreover, according to McKinsey, investment in the mobility and autonomous vehicle space has grown tremendously since the 2010s. With all this development, the range of automated vehicle technology has increased.
With so many products and solutions, how should we compare across them? For example, Ford’s BlueCruise technology provides a different level of autonomy than Cruise’s robotaxis. That’s where the Society of Automotive Engineers (SAE) International comes in — they are a group of 100K+ engineers and technical experts that develop standards and best practices. In 2014, they released SAE J3016™ Recommended Practice: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles, commonly referenced as the SAE Levels of Driving Automation™, which has served as the industry standard for describing automated vehicular solutions. The below table summarizes the SAE’s taxonomy.
In this article, we’ll discuss the levels of automated vehicles, from Level 0 (no driving automation) to Level 5 (full driving automation).
Terminology
Before diving into the levels, let’s define some of the key terms used by SAE.
Dynamic Driving Task (DDT): all the tactical and operational functions involved with driving and navigating traffic. This includes:
- lateral vehicle motion control which refers to steering (moving side to side)
- longitudinal vehicle motion control which refers to acceleration and deceleration (moving forward and backward)
- object and event detection and response (OEDR) which refers to monitoring the environment, detecting objects / events, and responding to them
- enabling conspicuity or communicating with other vehicles via signaling, lighting, sounding the horn, etc.
Note: DDT does not include the strategic aspects of driving, such as scheduling, trip planning, or determining the waypoints and stops.
Operational Design Domain (ODD): the conditions for where and when automation can operate. The factors to define the condition include, but are not limited to, environment, geography, time-of-day, and presence of certain roadway characteristics. For example, Ford’s BlueCruise features, which offers hands-free highway driving, is designed to operate on highways — thus the ODD for it are highways. Another example is Cruise’s robotaxis, which only function at night within the designated city perimeters.
Object and Event Detection and Response (OEDR): identifying objects and events involved with driving and executing the appropriate course of action. For example, if there’s another car in the way, it should be detected and then followed with a response; in this case, that might be moving to the adjacent lane. OEDR is a critical subtask within DDT. At lower levels of driving automation, OEDR is done by a human or the driver whereas at higher levels, its done by the vehicle / system.
DDT Fallback: achieving the minimum risk condition, which is a stable, stopped state where the vehicle is at a reduced risk of a crash, following either system failure, which are malfunctions or damages that render the vehicle or automated driving system unable to perform sustained driving, or exiting the ODD. This can be done by a human or the driver, which happens at lower levels of driving automation, or by the vehicle / system through a failure mitigation strategy.
Automated Driving System (ADS): the system, comprised of hardware and software, is capable of performing all parts of the DDT process on a sustained basis (within any ODD). The term is used to refer to solutions within Level 3–5.
In the definitions of OEDR and DDT Fallback, I use the phrase “done by a human or the driver,” drawing the distinction between human and driver involvement. Although the driver is human and encompassed in human involvement, vehicles are capable of being operated remotely, thus the human involved may not always be a driver. In SAE J3016, it’s stated that “Remote driving is not driving automation.” Moreover, in their 2021 version, they depreciate the usage of “unmanned” due to this nuance — vehicles can technically be unmanned while not being automated due to remote control (thus being unmanned doesn’t provide an accurate gauge on the level of automated driving).
Level 0: No Driving Automation
All components of DDT are done by a human. This includes navigating the road, OEDR, and DDT fallback. In this level, the concept of ODD is not defined because there’s no automation and all operations is done by the human operator.
Vehicles in this level can have active safety systems, which monitor conditions and potentially intervene to boost safety. Examples include vehicle maintenance alerts, collision detection, and lane departure warnings. Although these features maybe automated, they don’t engage in the driving process in a sustained fashion.
Most vehicles circa 2015 fall into this level.
Level 1: Driver Assistance
The automation system controls either the lateral (i.e., steering) or longitudinal (i.e., accelerating/decelerating) vehicle motion control (but not both) within their ODD, and the human performs the remaining DDT process. The human is responsible for OEDR and DDT Fallback. In terms of ODD scope, the automation system typically has a limited set of conditions its operational in.
Features within Level 1 driving automation are adaptive cruise control and lane assist. Adaptive Cruise Control (ACC) is when the vehicle controls its speed and automatically adjusts it to maintain a safe distance from the car ahead. With ACC, the automation system is controlling longitudinal vehicle motion. The human is still needed for lateral motion / steering the vehicle. In terms of ODD, ACC struggles to operate in bad weather conditions such as rain, fog, and snow as it can interfere with the sensors used. An example would be the Kia Stinger GT which has adaptive cruise control.
For lane assist, the driving automation monitors your location in the lane (through road markers), warns the human driver if its verging off, and can gently steer back into the lane. In this case, the automation is on the lateral motion. Similar to ACC, the feature doesn’t work in adverse weather or when the road markers aren’t clearly visible / detectable.
Level 2: Partial Driving Automation
In this level, the automation system controls both lateral (i.e., steering) or longitudinal (i.e., accelerating/decelerating) vehicle motion control within the ODD. The human is responsible for operating the vehicle when out of the ODD, performing OEDR at all times (inside and outside the ODD), and handling DDT Fallback. As such, the human driver needs to still be fully alert, monitor the road, and be ready to take over control.
Level 2 automation can be achieved by combining several features from Level 1. For example, combining adaptive cruise control and lane assist together to achieve both lateral and longitudinal driving automation. Other examples include automatic emergency braking (i.e., coming to a stop and avoiding crashes) and parking assistance (i.e., parking a car by managing acceleration / deceleration and steering). Ford’s BlueCruise technology would fall into this level.
Generally speaking, most modern / recent automated driving vehicles, as of writing this article, will fall into this level. Some examples are General Motors Super Cruise, Mercedes-Benz Distronic Plus, Nissan ProPilot Assist, and Tesla Autopilot. Furthermore, as of writing this article, this is the furthest level of driving automation commonly available to private / individual drivers.
Level 3: Conditional Driving Automation
This is when we achieve an automated driving system (ADS), as the system can perform lateral (i.e., steering) vehicle motion, longitudinal (i.e., accelerating/decelerating) vehicle motion, and OEDR within the ODD — the system does the full DDT process in the ODD. However, the human is still needed for DDT Fallback, when there’s a system failure or the vehicle exits the ODD.
Level 3 driving automation doesn’t fully remove the human, but it does remove the need for the human to be fully attentive while driving when in ODD. This is one of the biggest differences for the driver experience when comparing Level 2 and Level 3 vehicles. In Level 2, even when the vehicle automation is ongoing, the driver needs to be fully alert as they are responsible for OEDR. However, in Level 3, the human is not needed while in the ODD — they just need to be ready when DDT Fallback occurs.
Currently, there are very few vehicles or products that fall into Level 3 as of writing this article. At the moment, Mercedes-Benz’s DRIVE PILOT technology is an example of Level 3 automation. In early 2023, Mercedes-Benz got approval from the state of Nevada to deploy their ADS on the road, making them one of the first to bring Level 3 vehicles to the US.
Level 4: High Driving Automation
With Level 4 automation, the ADS manages both the DDT process (all components and subtasks) and DDT Fallback within the ODD. In this level, there’s no expectation of a human / driver intervening while the vehicle is in the ODD. Thus, unlike Level 3, where a human might still need to be present to handle the DDT fallback, the human can potentially be removed altogether (while operating in the ODD). The limitation of Level 4 ADS is that it doesn’t operate outside its ODD.
As of writing this article, there are only a handful of Level 4 vehicles out on the road, being used in a real-world setting (however, many automakers and technology companies are working on creating them). For example, there is Waymo and Cruise, which have their vehicles deployed in various cities in the US. For Waymo and Cruise in particular, both these companies offer robotaxi solutions that can remove the human in the loop (source: Waymo in San Francisco, Waymo in Phoenix, Cruise in San Francisco, and Cruise in Phoenix). It should be noted that their solutions are designed for their ODD, which are the cities they currently operate in (such as San Francisco and Phoenix). Furthermore, for Cruise, their service only operates at certain times and certain speeds (thus a more restrictive ODD).
For Waymo and Cruise, the general approach they have taken to Level 4 automation is to use LiDAR, camera, RADAR, and other sensors to create a map of their environment. With an accurate map of the environment, which are usually updated with every drive, the vehicle can anticipate common driving scenarios and navigate the roads. Thus, this “map” will be one of the key components and limiting factors defining the vehicle’s ODD. In general, vehicles with this approach are geofenced. For example, although Waymo’s / Cruise’s vehicles operate well in their current cities, if it was placed in a new location / city, they wouldn’t operate effectively as they don’t have an accurate map of its environment.
An alternative approach is to rely more on general computer vision and sensor fusion, where the vehicle responds to real-time camera and sensor input. Thus, instead of relying on a pre-established map of the environment, the vehicle adapts based on real time measurements. This is currently the approach being pursued at Tesla (as of writing this article). A key limitation of this approach is the complexity of driving and making real-time decisions based on real-time data.
Level 5: Full Automation
This is the last stage of driving automation — the ADS does the full DDT process and DDT fallback in every condition. For Level 5, the ODD is unlimited as the vehicle is expected to operate every and anywhere. With this level, the human is completely out of the loop when it comes to driving (they may still be involved in maintenance, trip planning, scheduling, etc.).
As of writing this article, there are no Level 5 solutions, however many companies and organizations are working towards it. Some example are Waymo, Cruise, Nuro, Uber, and Toyota. The realization of Level 5 vehicles can potentially have huge implications on how society functions. When Level 5 technology will be achieved, if it ever will be achieved, is still a mystery (as of writing this article). Some believe it will come into fruition in 10–20 years while others argue it isn’t possible.
Conclusion
Over the past several decades, a number of automated solutions for driving have been developed and deployed. The SAE has developed standards to better classify the variety of technologies in the automated driving space. In this article, we explored the terminology and levels of driving automation as described in their J3016 (2021) standard.
