The Taxonomy for Ground Vehicles Levels of Automation 

cockpit of autonomous car. a vehicle running self driving mode and a woman driver reading book.

The automotive world is on the verge of a revolution with vehicles reaching higher levels of automation. Automation is defined by the Oxford English Dictionary as: “the use of electronic or mechanical devices to replace human labor”. Among all the existing concepts (e.g., Google, Uber, Tesla, but also Renault, Toyota and many more) the automation level can vary greatly. Indeed, different concepts have different goals, some of them aim at removing the human from the dynamical parts of the driving task, while others want to keep the human “in” or “on” the loop (e.g., cars with or without steering wheels). This created the need for a classification of the different goals and levels of automation, which led to the birth of the “J3016: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles” document (can be bought obtained for free here). This document is the result of a common US/Europe effort to create a descriptive and informative but not normative document, which writing has been entrusted to the SAE on-road automated vehicle standards committee.

The main contribution of this document is the determination of six levels of automation, along with precise definitions for several terms.

 

This document divides the act of driving in three stages: strategic (e.g., choosing the route, timing), tactical (e.g., motion inside the traffic), and operational (e.g. reflex reactions). It is pointed out that the automation levels considered do not deal with the strategic part. Numerous definitions are provided, here are a few of the most useful ones with simple definitions of our own (see the actual document for precise definitions):

  • Dynamic Driving Task (DDT): what most people would call driving, involves the tactical and operational parts mentioned earlier.
  • Dynamic Driving Task Fallback (DDT fallback): the system that deals with the situation when things go wrong. Can be the human itself.
  • Automated Driving System (ADS): the automation part of the car (hardware & software).
  • Operational Design Domain (ODD): specific operational description in which an ADS will be designed to operate.

These four terms are the most used throughout the document, and in the main tables/figures. With these terms being defined, the document goes on to describe the six levels of automation that an ADS can provide. These are described in the following table.

taxonomy_cars_autonomy

Basically,

– Level 0 = not automation at all.

– Level 1 = longitudinal or lateral distance handled autonomously.

– Level 2 = both distances handled, but human must continuously monitor the system.

– Level 3 = human does not need to monitor the system but he is the fallback system (must intervene if something wrong happens).

– Level 4 = human not needed anymore (even for fallback), but operations subjected to some limitations.

– Level 5 = like level 4 but with no operational limitations.

These six levels allow to describe precisely the levels of autonomy of existing cars and of the ones to come, for example Tesla cars are Level 2. Some more interesting concepts are detailed through the document, like the possibility to have a remote driver, or a dispatcher capable of initializing or deactivate the ADS. The monitoring (permanent vigilance) and receptive (no vigilance but can be alerted) concepts are detailed. The notion of minimal risk condition, some sort of contingency procedure, is also discussed. And more notions that would not fit in this short article.

 

To conclude, this document is not a standard, as it is repeated throughout the text; it is a descriptive and informative document. Yet, it is very detailed and clear, with numerous examples to illustrate the various concepts. From all the concepts, it appears that a parallel with aeronautics could be done, for example a “minimal risk condition” looks a lot like a “contingency procedure”. Defining a dictionary of equivalent notion to translate notion from one world to the other would appear highly useful. Especially as drone levels of automation are not yet clearly categorized and could use such a taxonomy, at least as a starting basis.

Leave a Reply