Primer on SAE Levels of Autonomy

[diplomat33]

Since I know we have a lot of discussions about the SAE levels, I thought I would write up a primer from the SAE document:

General

• The levels are not in order of quality, or merit or technology. L5 is not necessarily “better” than L4. L4 is not necessarily “better” than L3.
• A “high functioning” L3 does not become a “low functioning” or “partial” L4. It is still L3.
• SAE levels are never fractions. For example, if a system qualifies as L3 system, it is L3, not L3.5 or L3.7.
• The SAE levels are discrete and mutually exclusive. So a system or feature cannot be both L3 and L4. It is one or the other.
• It is possible for a car to have different features of different levels. For example, our Teslas have TACC which is L1 and AP which is L2. When AP is on, our cars operate at L2 and when TACC is on, they operate at L1.
• The SAE does not expect autonomous driving systems to do strategic tasks like trip planning or scheduling that are typically done before starting a trip. The SAE levels only expect the autonomous driving to perform dynamic driving tasks (DDT) which are the driving tasks performed during the actual driving part.
• The SAE levels are defined based on the role that the human and the automated driving system have during a trip.
• Active safety systems like automatic emergency braking or emergency lane departure avoidance can be included on cars of any levels.
• Individual features can have a level of autonomy. For example, auto park with no driver inside could be a L4 feature.

Monitoring human attention

• Only L2 and L3 need to monitor the human since L2 needs the human to be an attentive driver and L3 may need the human to intervene.
• L4 and L5 do not need to monitor human attention since the systems are capable of pulling over in case of a problem on their own.

Monitoring vehicle and system failure

• L3, L4 and L5 need to be able to self-monitor their systems for any problems.
• For L1 and L2, the human is expected to monitor the vehicle and system for any problems.

Object, Event, Detection, Response (OEDR)

• A sub task of DDT that involves detecting, recognizing, and classifying objects and events, preparing a response and executing a response. Example: seeing a stop sign and stopping in time.
• If the system can do entire OEDR in its ODD, then it is L3-L5 autonomous, If it cannot do entire OEDR, then it is a “driver assist” (L1-L2).

Operational Design Domain (ODD)

• The conditions that an automated driving system can operate in. Conditions include environmental, geographical, time of day restrictions, presence of absence of traffic or roadway characteristics.
• If the ODD has special restrictions due to technological limits, then it is L3/L4. For example, limited to low speeds or limited to only highways because the automated driving system cannot operate safely outside those limits.
• L5 ODD is defined as all driver-managebale road conditions within its region of the world. For example, L5 is not expected to operate in flooded roads, or white out snow storms.
• Non technological limits to the ODD don’t count. For example, an autonomous vehicle that could operate in Canada but is geofenced to the US due to legal reasons, would still be considered L5.
• Example of L5 ODD: a robotaxi that can operate on all driver manageable roads in the US, day or night, all weather that human can handle, all legal speeds.

Receptivity of human

• L0-L2, the human is expected to fully receptive to what is going on.
• L3, the human should be receptive to a vehicle failure or the system requesting intervention.

Supervision of driving automation system

• For L1 or L2, the human is expected to supervise the system.

Sustained operation

• The SAE levels require that the driving feature be sustainable, meaning that they must be able to continue functioning across external events. Example, adaptive cruise control counts because it can change the speed based on change in speed of the front car. But dumb cruise control does not count as sustained operation since it cannot reacting to change in front car’s speed.

Usage specification

• The SAE level for a particular ODD.
• Since L3 or L4 system can have different limited ODD, it is necessary to specify both the SAE level and the ODD in order to fully describe the capabilities of a L3 or L4 system or feature.
• Since L5 has all encompassing ODD, it is not necessary to specify ODD for L5.

Level 0: No automation

• Driving Automation System (if any): Does not perform any part of the DDT on a sustained basis (although other vehicle systems may provide warnings or support, such as momentary emergency intervention).
• Human (at all times): Performs the entire DDT
• For levels 1-2, the automated driving system performs only part of the DDT.

For level 1-2, the driver automated system only performs part of the DDT

Level 1: Driver Assist

• Driving Automation System (while engaged): Performs part of the DDT by executing either the longitudinal or the lateral vehicle motion control subtask. Disengages immediately upon driver request
• Human (at all times): Performs the remainder of the DDT not performed by the driving automation system. Supervises the driving automation system and intervenes as necessary to maintain safe operation of the vehicle. Determines whether/when engagement or disengagement of the driving automation system is appropriate. Immediately performs the entire DDT whenever required or desired.

Level 2: partial driving automation

• Driving Automation System (while engaged): Performs part of the DDT by executing both the lateral and the longitudinal vehicle motion control subtasks. Disengages immediately upon driver request.
• Human (at all times): Performs the remainder of the DDT not performed by the driving automation system. Supervises the driving automation system and intervenes as necessary to maintain safe operation of the vehicle. Determines whether/when engagement and disengagement of the driving automation system is appropriate. Immediately performs the entire DDT whenever required or desired.
• L2 can bring the car to a controlled stop in the middle of the lane if the driver fails to supervise the feature. For example, AP does this now when it comes to a controlled stop if we ignore the nags long enough.

For levels 3-5, the automated driving system performs the entire DDT.

Level 3: conditional driving automation

• ADS (while not engaged): Permits engagement only within its ODD
• ADS (while engaged): Performs the entire DDT. Determines whether ODD limits are about to be exceeded and, if so, issues a timely request to the human to intervene. Determines whether there is a DDT performance-relevant system failure of the ADS and, if so, issues a timely request to the human to intervene. Disengages an appropriate time after issuing a request to intervene. Disengages immediately upon driver request
• Human (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines when engagement of ADS is appropriate. Becomes the DDT fallback-ready user when the ADS is engaged
• Human (while the ADS is engaged): Is receptive to a request to intervene and responds by performing DDT fallback in a timely manner. Is receptive to DDT performance-relevant system failures in vehicle systems and, upon occurrence, performs DDT fallback in a timely manner. Determines whether and how to achieve a minimal risk condition. Becomes the driver upon requesting disengagement of the ADS
• L3 can bring the car to a controlled stop in the middle of the lane if the driver fails to take over when prompted.

Examples of L3 fallback:

• L3 system is engaged and experiences a vehicle failure (ie flat tire). The human recognizes failure and takes over and pulls the car over to the side of the road.
• L3 system is engaged and experiences a system failure (ie sensors are obstructed). The L3 system notifies the human who takes over and resumes manual driving.
• L3 system is engaged but is about to leave the conditions that it can operate in (for ex: leaving the highway). It notifies the human who takes over and resumes manual driving.

Level 4: High driving automation

• ADS (while not engaged): Permits engagement only within its ODD
• ADS (while engaged): Performs the entire DDT. May issue a timely request to intervene. Performs DDT fallback and transitions automatically to a minimal risk condition when: A DDT performance-relevant system failure occurs or a user does not respond to a request to intervene or a user requests that it achieve a minimal risk condition. Disengages, if appropriate, only after: it achieves a minimal risk condition or a driver is performing the DDT. May delay user-requested disengagement.
• Driver/dispatcher (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines whether to engage the ADS. Becomes a passenger when the ADS is engaged only if physically present in the vehicle.
• Passenger/dispatcher (while the ADS is engaged): Need not perform the DDT or DDT fallback. Need not determine whether and how to achieve a minimal risk condition. May perform the DDT fallback following a request to intervene. May request that the ADS disengage and may achieve a minimal risk condition after it is disengaged. May become the driver after a requested disengagement.
• In a emergency like a total loss of power that would not give the car enough time to pull over, L4 can bring the car to a controlled stop in the lane.

Level 5: Full driving automation

• ADS (while not engaged): Permits engagement of the ADS under all driver-manageable on-road conditions
• ADS (while engaged): Performs the entire DDT. Performs DDT fallback and transitions automatically to a minimal risk condition when: a DDT performance-relevant system failure occurs or a user does not respond to a request to intervene or a user requests that it achieve a minimal risk condition. Disengages, if appropriate, only after: It achieves a minimal risk condition or a driver is performing the DDT. May delay a user-requested disengagement
• Driver/dispatcher (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines whether to engage the ADS. Becomes a passenger when the ADS is engaged only if physically present in the vehicle
• Passenger/dispatcher (while the ADS is engaged): Need not perform the DDT or DDT fallback. Need not determine whether and how to achieve a minimal risk condition. May perform the DDT fallback following a request to intervene. May request that the ADS disengage and may achieve a minimal risk condition after it is disengaged. May become the driver after a requested disengagement.
• In an emergency like a total loss of power that would not give the car enough time to pull over, L5 can bring the car to a controlled stop in the lane.

Hope this helps.

 

[jebinc]

Since I know we have a lot of discussions about the SAE levels, I thought I would write up a primer from the SAE document:

General

• The levels are not in order of quality, or merit or technology. L5 is not necessarily “better” than L4. L4 is not necessarily “better” than L3.
• A “high functioning” L3 does not become a “low functioning” or “partial” L4. It is still L3.
• SAE levels are never fractions. For example, if a system qualifies as L3 system, it is L3, not L3.5 or L3.7.
• The SAE levels are discrete and mutually exclusive. So a system or feature cannot be both L3 and L4. It is one or the other.
• It is possible for a car to have different features of different levels. For example, our Teslas have TACC which is L1 and AP which is L2. When AP is on, our cars operate at L2 and when TACC is on, they operate at L1.
• The SAE does not expect autonomous driving systems to do strategic tasks like trip planning or scheduling that are typically done before starting a trip. The SAE levels only expect the autonomous driving to perform dynamic driving tasks (DDT) which are the driving tasks performed during the actual driving part.
• The SAE levels are defined based on the role that the human and the automated driving system have during a trip.
• Active safety systems like automatic emergency braking or emergency lane departure avoidance can be included on cars of any levels.
• Individual features can have a level of autonomy. For example, auto park with no driver inside could be a L4 feature.

Monitoring human attention

• Only L2 and L3 need to monitor the human since L2 needs the human to be an attentive driver and L3 may need the human to intervene.
• L4 and L5 do not need to monitor human attention since the systems are capable of pulling over in case of a problem on their own.

Monitoring vehicle and system failure

• L3, L4 and L5 need to be able to self-monitor their systems for any problems.
• For L1 and L2, the human is expected to monitor the vehicle and system for any problems.

Object, Event, Detection, Response (OEDR)

• A sub task of DDT that involves detecting, recognizing, and classifying objects and events, preparing a response and executing a response. Example: seeing a stop sign and stopping in time.
• If the system can do entire OEDR in its ODD, then it is L3-L5 autonomous, If it cannot do entire OEDR, then it is a “driver assist” (L1-L2).

Operational Design Domain (ODD)

• The conditions that an automated driving system can operate in. Conditions include environmental, geographical, time of day restrictions, presence of absence of traffic or roadway characteristics.
• If the ODD has special restrictions due to technological limits, then it is L3/L4. For example, limited to low speeds or limited to only highways because the automated driving system cannot operate safely outside those limits.
• L5 ODD is defined as all driver-managebale road conditions within its region of the world. For example, L5 is not expected to operate in flooded roads, or white out snow storms.
• Non technological limits to the ODD don’t count. For example, an autonomous vehicle that could operate in Canada but is geofenced to the US due to legal reasons, would still be considered L5.
• Example of L5 ODD: a robotaxi that can operate on all driver manageable roads in the US, day or night, all weather that human can handle, all legal speeds.

Receptivity of human

• L0-L2, the human is expected to fully receptive to what is going on.
• L3, the human should be receptive to a vehicle failure or the system requesting intervention.

Supervision of driving automation system

• For L1 or L2, the human is expected to supervise the system.

Sustained operation

• The SAE levels require that the driving feature be sustainable, meaning that they must be able to continue functioning across external events. Example, adaptive cruise control counts because it can change the speed based on change in speed of the front car. But dumb cruise control does not count as sustained operation since it cannot reacting to change in front car’s speed.

Usage specification

• The SAE level for a particular ODD.
• Since L3 or L4 system can have different limited ODD, it is necessary to specify both the SAE level and the ODD in order to fully describe the capabilities of a L3 or L4 system or feature.
• Since L5 has all encompassing ODD, it is not necessary to specify ODD for L5.

Level 0: No automation

• Driving Automation System (if any): Does not perform any part of the DDT on a sustained basis (although other vehicle systems may provide warnings or support, such as momentary emergency intervention).
• Human (at all times): Performs the entire DDT
• For levels 1-2, the automated driving system performs only part of the DDT.

For level 1-2, the driver automated system only performs part of the DDT

Level 1: Driver Assist

• Driving Automation System (while engaged): Performs part of the DDT by executing either the longitudinal or the lateral vehicle motion control subtask. Disengages immediately upon driver request
• Human (at all times): Performs the remainder of the DDT not performed by the driving automation system. Supervises the driving automation system and intervenes as necessary to maintain safe operation of the vehicle. Determines whether/when engagement or disengagement of the driving automation system is appropriate. Immediately performs the entire DDT whenever required or desired.

Level 2: partial driving automation

• Driving Automation System (while engaged): Performs part of the DDT by executing both the lateral and the longitudinal vehicle motion control subtasks. Disengages immediately upon driver request.
• Human (at all times): Performs the remainder of the DDT not performed by the driving automation system. Supervises the driving automation system and intervenes as necessary to maintain safe operation of the vehicle. Determines whether/when engagement and disengagement of the driving automation system is appropriate. Immediately performs the entire DDT whenever required or desired.
• L2 can bring the car to a controlled stop in the middle of the lane if the driver fails to supervise the feature. For example, AP does this now when it comes to a controlled stop if we ignore the nags long enough.

For levels 3-5, the automated driving system performs the entire DDT.

Level 3: conditional driving automation

• ADS (while not engaged): Permits engagement only within its ODD
• ADS (while engaged): Performs the entire DDT. Determines whether ODD limits are about to be exceeded and, if so, issues a timely request to the human to intervene. Determines whether there is a DDT performance-relevant system failure of the ADS and, if so, issues a timely request to the human to intervene. Disengages an appropriate time after issuing a request to intervene. Disengages immediately upon driver request
• Human (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines when engagement of ADS is appropriate. Becomes the DDT fallback-ready user when the ADS is engaged
• Human (while the ADS is engaged): Is receptive to a request to intervene and responds by performing DDT fallback in a timely manner. Is receptive to DDT performance-relevant system failures in vehicle systems and, upon occurrence, performs DDT fallback in a timely manner. Determines whether and how to achieve a minimal risk condition. Becomes the driver upon requesting disengagement of the ADS
• L3 can bring the car to a controlled stop in the middle of the lane if the driver fails to take over when prompted.

Examples of L3 fallback:

• L3 system is engaged and experiences a vehicle failure (ie flat tire). The human recognizes failure and takes over and pulls the car over to the side of the road.
• L3 system is engaged and experiences a system failure (ie sensors are obstructed). The L3 system notifies the human who takes over and resumes manual driving.
• L3 system is engaged but is about to leave the conditions that it can operate in (for ex: leaving the highway). It notifies the human who takes over and resumes manual driving.

Level 4: High driving automation

• ADS (while not engaged): Permits engagement only within its ODD
• ADS (while engaged): Performs the entire DDT. May issue a timely request to intervene. Performs DDT fallback and transitions automatically to a minimal risk condition when: A DDT performance-relevant system failure occurs or a user does not respond to a request to intervene or a user requests that it achieve a minimal risk condition. Disengages, if appropriate, only after: it achieves a minimal risk condition or a driver is performing the DDT. May delay user-requested disengagement.
• Driver/dispatcher (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines whether to engage the ADS. Becomes a passenger when the ADS is engaged only if physically present in the vehicle.
• Passenger/dispatcher (while the ADS is engaged): Need not perform the DDT or DDT fallback. Need not determine whether and how to achieve a minimal risk condition. May perform the DDT fallback following a request to intervene. May request that the ADS disengage and may achieve a minimal risk condition after it is disengaged. May become the driver after a requested disengagement.
• In a emergency like a total loss of power that would not give the car enough time to pull over, L4 can bring the car to a controlled stop in the lane.

Level 5: Full driving automation

• ADS (while not engaged): Permits engagement of the ADS under all driver-manageable on-road conditions
• ADS (while engaged): Performs the entire DDT. Performs DDT fallback and transitions automatically to a minimal risk condition when: a DDT performance-relevant system failure occurs or a user does not respond to a request to intervene or a user requests that it achieve a minimal risk condition. Disengages, if appropriate, only after: It achieves a minimal risk condition or a driver is performing the DDT. May delay a user-requested disengagement
• Driver/dispatcher (while the ADS is not engaged): Verifies operational readiness of the ADS-equipped vehicle. Determines whether to engage the ADS. Becomes a passenger when the ADS is engaged only if physically present in the vehicle
• Passenger/dispatcher (while the ADS is engaged): Need not perform the DDT or DDT fallback. Need not determine whether and how to achieve a minimal risk condition. May perform the DDT fallback following a request to intervene. May request that the ADS disengage and may achieve a minimal risk condition after it is disengaged. May become the driver after a requested disengagement.
• In an emergency like a total loss of power that would not give the car enough time to pull over, L5 can bring the car to a controlled stop in the lane.

Hope this helps.

@diplomat33
This is a great summary and should help many understand the details, should they decide not to invest the time reading the entire SAE Autonomy document. Thanks for taking the time to pull this together!

Edit: Down the road (r1), maybe define/expand the "ODD" acronym, before using the "ODD" acronym (in general, make sure all acronyms are defined/expanded, before (or while) using them), and provide some more real world "Tesla" examples, that members can relate to, in each "Level" description. Thanks again for doing this!

 

[Bladerskb]

A few points i feel were omitted or not expounded enough on which I would like to add because I feel they are paramount. Other than that good summary!

Level 3 Handover

At level 3, an ADS is capable of continuing to perform the DDT for at least several seconds after providing the fallback-ready user with a request to intervene. The DDT fallback-ready user is then expected to achieve a minimal risk condition if s/he determines it to be necessary. ​

Monitoring

When an ADS is on and is likewise performing the entire DDT the driver CANNOT be required to pay attention in-order to monitor the environment (road) or the system. Therefore a system you have to monitor or a system that requires you to pay attention, no matter how good it is. It is not a ADS (levels 3-5)!

When operating conventional vehicles that are not equipped with an engaged ADS, drivers visually sample the road scene sufficiently to competently perform the DDT while also performing secondary tasks that require short periods of eyes-off-road time (e.g., adjusting cabin comfort settings, scanning road signs, tuning a radio, etc.). Thus, monitoring the driving environment does not necessarily entail continuous eyes on-road time by the driver.

At levels 1-2, the driver monitors the driving automation system’s performance.
At higher levels of driving automation (levels 3-5), the ADS monitors its own performance of the complete DDT.
​

Receptive User vs Monitoring in Levels 0-2

The driver state or condition of being receptive to alerts or other indicators of a DDT performance-relevant system failure, as assumed in level 3, is not a form of monitoring. The difference between receptivity and monitoring is best illustrated by example: A person who becomes aware of a fire alarm or a telephone ringing may not necessarily have been monitoring the fire alarm or the telephone. Likewise, a user who becomes aware of a trailer hitch falling off may not necessarily have been monitoring the trailer hitch. By contrast, a driver in a vehicle with an active level 1 adaptive cruise control (ACC) system is expected to monitor both the driving environment and the ACC performance and otherwise not to wait for an alert to draw his/her attention to a situation requiring a response​

 

[diplomat33]

A few points i feel were omitted or not expounded enough on which I would like to add because I feel they are paramount. Other than that good summary!

Level 3 Handover

At level 3, an ADS is capable of continuing to perform the DDT for at least several seconds after providing the fallback-ready user with a request to intervene. The DDT fallback-ready user is then expected to achieve a minimal risk condition if s/he determines it to be necessary. ​

Monitoring

When an ADS is on and is likewise performing the entire DDT the driver CAN'T be required to pay attention in-order to monitor the environment (road) or the system. Therefore a system you have to monitor or require you to pay attention, no matter how good it is. It is not a ADS (L3, L4 or L5)!

When operating conventional vehicles that are not equipped with an engaged ADS, drivers visually sample the road scene sufficiently to competently perform the DDT while also performing secondary tasks that require short periods of eyes-off-road time (e.g., adjusting cabin comfort settings, scanning road signs, tuning a radio, etc.). Thus, monitoring the driving environment does not necessarily entail continuous eyes on-road time by the driver.

At levels 1-2, the driver monitors the driving automation system’s performance.
At higher levels of driving automation (levels 3-5), the ADS monitors its own performance of the complete DDT.
​

Receptive User vs Monitoring in L0, L1, and L2

The driver state or condition of being receptive to alerts or other indicators of a DDT performance-relevant system failure, as assumed in level 3, is not a form of monitoring. The difference between receptivity and monitoring is best illustrated by example: A person who becomes aware of a fire alarm or a telephone ringing may not necessarily have been monitoring the fire alarm or the telephone. Likewise, a user who becomes aware of a trailer hitch falling off may not necessarily have been monitoring the trailer hitch. By contrast, a driver in a vehicle with an active level 1 adaptive cruise control (ACC) system is expected to monitor both the driving environment and the ACC performance and otherwise not to wait for an alert to draw his/her attention to a situation requiring a response​

Yeah, I am sure I missed a few things. So thank you for adding that.

 

[diplomat33]

Edit: Down the road (r1), maybe define/expand the "ODD" acronym, before using the "ODD" acronym (in general, make sure all acronyms are defined/expanded, before (or while) using them), and provide some more real world "Tesla" examples, that members can relate to, in each "Level" description. Thanks again for doing this!

Thanks for the tips. It took me a long time to write the summary. I went through several drafts. So I am sure the summary is not perfect. Feedback is welcomed.

In terms of more Tesla examples, The current NOA alert when you are about to leave the highway seems to fit a L3 fallback feature. It notifies the driver to take over when it is about to leave it's ODD and gives the driver a few seconds to take over. It also smoothly transitions down to L2 AP.

Since I can't edit the OP anymore, and you asked for more info on ODD, let me add some more below:

Significance of ODD

• It is possible to have different ODDs for the same SAE level (L1-L4). It depends on how the manufacturer designs the autonomous driving system or feature. Example: one manufacturer may design adaptive cruise control that only works at low speeds while another manufacturer designs adaptive cruise control that works for all speeds.
• Because of the wide range of possible ODDs, a wide range of possible features may exist in each level (e.g., level 4 includes parking, high-speed, low-speed, geo-fenced, etc.)
• ODD is particularly important in explaining why a system is not L5 since by definition, any limitations in the ODD is what makes the system "not L5". Example: if a system is limited to only limited access divided highways, then you know that the reason the system is not L5 is because it only works on limited access divided highways.
• A given driving automation system feature has only one ODD, but that ODD may be quite varied and multi-faceted. Even though the ODD is composed of multiple variables, it would be incorrect to say that a driving automation feature has multiple ODDs.

 

[jebinc]

Thanks for the tips. It took me a long time to write the summary. I went through several drafts. So I am sure the summary is not perfect. Feedback is welcomed.

In terms of more Tesla examples, The current NOA alert when you are about to leave the highway seems to fit a L3 fallback feature. It notifies the driver to take over when it is about to leave it's ODD and gives the driver a few seconds to take over. It also smoothly transitions down to L2 AP.

Since I can't edit the OP anymore, and you asked for more info on ODD, let me add some more below:

Significance of ODD

• It is possible to have different ODDs for the same SAE level (L1-L4). It depends on how the manufacturer designs the autonomous driving system or feature. Example: one manufacturer may design adaptive cruise control that only works at low speeds while another manufacturer designs adaptive cruise control that works for all speeds.
• Because of the wide range of possible ODDs, a wide range of possible features may exist in each level (e.g., level 4 includes parking, high-speed, low-speed, geo-fenced, etc.)
• ODD is particularly important in explaining why a system is not L5 since by definition, any limitations in the ODD is what makes the system "not L5". Example: if a system is limited to only limited access divided highways, then you know that the reason the system is not L5 is because it only works on limited access divided highways.
• A given driving automation system feature has only one ODD, but that ODD may be quite varied and multi-faceted. Even though the ODD is composed of multiple variables, it would be incorrect to say that a driving automation feature has multiple ODDs.

@diplomat33
Thanks again for taking the time to summarize! I’m sure many will find it helpful!

 

[S4WRXTTCS]

That's a great job summarizing them.

I figured if anyone would try to summarize them it would be you, and to be honest I don't understand your obsession with them. They're great from an historical perspective on how we thought autonomous driving was going to go, but in reality they're not much use these days.

These days I mostly just mock them.

SAE Fantasy L1 -> Allow for either lane-steering or adaptive cruise control, but not both,.
Reality L1 -> Why would anyone do steering without adaptive cruise control? Oh, and the whole not requiring driving monitoring for adaptive cruise control didn't go so well with a society addicted to their cell phones. Any excuse to text is an excuse they'll use,.

SAE Fantasy L2 -> Allow for both lane-steering, and adaptive cruise control. Lets say it requires driver monitoring, but we're not going to say how or that it even has to be done well. We're also not going to put restrictions on how automated you can go while also still requiring human oversight even though it's been demonstrated time, and time again that humans can't oversee a highly automated task.
Reality L2 -> Lots and lots of police cars and fire trucks have been crashed into because of poor drive monitoring systems on L2 vehicles (especially with the Tesla emblem on them).

SAE Fantasy L3 -> Self-driving, but allow the manufacture to define the ODD. We won't define how well the driver monitoring has to be, and we'll completely forget about a persons tendency to fall asleep while being a passenger.
Reality L3 -> What Level 3? There is only one L3 car trying to get approval, but only in Germany. And, that car has some significant ODD limits to it's L3 capabilities. Which are allowed by the SAE Level 3, but seriously limit its usefulness.

SAE L4 -> We've done a lousy job defining three levels of automation so we should a better job on this one.
Reality L4 -> Pretty solid. Gives lots of room for improvements through expanding geo-fenced areas, and ODD capabilities.

SAE Fantasy L5 -> We're going to fantasize a lot on this one. Let's start by declaring all human drivers capable of the same ODD's, and make that the requirement.
Reality L5 -> Maybe next decade, maybe?

On a more serious note only two levels even apply to a new Tesla. Those levels are L2, and L4 where you if you don't buy FSD then you're stuck with L2. If you do buy FSD you might or might not eventually get an L4 experience with it.

L0, and L1 doesn't apply to Tesla because they all come with L2, and no one cares about L0 to L1 distinction.
L3 doesn't apply to Tesla because HW3 vehicles don't have the proper driver monitoring for that.
L5 doesn't apply to Tesla because HW3 can't handle the ODD situations that even the most average of average drivers can easily handle.

 

[Daniel in SD]

I figured if anyone would try to summarize them it would be you, and to be honest I don't understand your obsession with them. They're great from an historical perspective on how we thought autonomous driving was going to go, but in reality they're not much use these days.

The SAE taxonomy and definitions are important because they’re the basis for current regulation of autonomous vehicles. Tesla says it plans to make autonomous vehicles that operate on public roads.

SAE Fantasy L1 -> Allow for either lane-steering or adaptive cruise control, but not both,.
Reality L1 -> Why would anyone do steering without adaptive cruise control? Oh, and the whole not requiring driving monitoring for adaptive cruise control didn't go so well with a society addicted to their cell phones. Any excuse to text is an excuse they'll use,.

Do you have reference for this? The SAE document says this has been studied and L1 systems are not being abused.

Anyway it seems like your main complaint is that progress has been slow.

 

[Bladerskb]

I figured if anyone would try to summarize them it would be you, and to be honest I don't understand your obsession with them. They're great from an historical perspective on how we thought autonomous driving was going to go, but in reality they're not much use these days.

Yeah like Daniel said, all regulation and laws not just in the US are based on the SAE definitions. They use the definitions to make the laws. So not totally useless.

Infact Japan just passed a law that goes into effect this year that legalizes Level 3 automation.

Here's a snippet:

Similar to other countries that have set goals to introduce self-driving vehicles, the Japanese Government adopted the Society of Automotive Engineers (SAE) J3016 Levels of Driving Automation1 (SAE Automation Levels) in order to describe the levels of driving automation that Japan intends to adopt. ​

Legalization of Self-Driving Vehicles in Japan: Progress Made, but Obstacles Remain - Lexology
Japan revamps laws to put self-driving cars on roads

 

[S4WRXTTCS]

Do you have reference for this? The SAE document says this has been studied and L1 systems are not being abused.

We know for a fact that L2 is being abused, but we really only know that because the first thing that comes up with a Tesla accident is "was AP on?". When Senators talk about banning AP they never even think of the millions of other cars on the road that have either a full L2 system or a L1 system plus a safety based lane-keeping (which is different than lane-steering and I don't believe qualifies the operation as L2).

I have very little doubt that L2 systems and L1 systems (with and without lane-keeping) are being used an excuse to drive distracted.

Here is the latest I've read on the topic.
New study: Adaptive cruise-control and other driver ...https://www.seattletimes.com › seattle-news › transportation › new-study-ada...

Anyway it seems like your main complaint is that progress has been slow.

I don't think SAE levels have anything to do with progress being slow. My main complaint about SAE levels is they don't reflect the real reality of autonomous driving. Sure it's good historical data, and it's good reference information for regulators in writing laws.

But, it's of very little actual information value especially to Tesla owners.

The problem is the SAE levels leave a lot up to interpretation, and to regulators to define what's allowed and not allowed. That's why an L2 system in Europe is so much more constrained than an L2 systems in the US.

Both you, and I likely completely agree on what's going to happen this year with Tesla. What's going to happen this year is Tesla is going to push the boundaries of what an L2 system is allowed to do. Where the NHTSA is going to continue doing their investigations every time some incident happens.

The SAE Level 2 doesn't constrict Tesla on what they can do with it. They can implement the entire OEDR while still refusing to allow the car to be L4. The regulators on the other hand do care, and they will nerf it if they feel like if its unsafe or if its asking too much for the human to oversee it.

The other reasons the SAE Levels don't really matter much to a Tesla FSD owner is that Tesla themselves don't care about the SAE Levels. If they did the entire game plan would be different.

Instead when asked Elon tried to claim that the system was going to be L5 which we both know is completely laughable.

 

[S4WRXTTCS]

Yeah like Daniel said, all regulation and laws not just in the US are based on the SAE definitions. They use the definitions to make the laws. So not totally useless.

Infact Japan just passed a law that goes into effect this year that legalizes Level 3 automation.

Here's a snippet:

Similar to other countries that have set goals to introduce self-driving vehicles, the Japanese Government adopted the Society of Automotive Engineers (SAE) J3016 Levels of Driving Automation1 (SAE Automation Levels) in order to describe the levels of driving automation that Japan intends to adopt. ​

Legalization of Self-Driving Vehicles in Japan: Progress Made, but Obstacles Remain - Lexology
Japan revamps laws to put self-driving cars on roads

But, we both know Tesla can't do L3. It's why you don't even own a Tesla. It's also why made sure to make sure the Driver Monitoring elements of the summary were solid..

Japan is an interesting because it's one area of the world where I could see a L3 driving being plausible. The culture itself is a lot different than the US.

I just can't see L3 being realistic for the US market. I think we're just going to skip right past that.

In the US market we really only have two levels of automation that anyone cares about. Those are L2, and L4.

 

[S4WRXTTCS]

For the sake of completeness I do want to bring up a few good reasons for referring to the SAE Levels.

It's common in the media to refer to advanced L2 systems as L2+, and I think its important to try to squash these down to L2 systems to make sure people don't get confused. In the OP's summary it makes it clear that an L2 system is still an L2 system regardless of how advanced it is.

It's common especially among Tesla owners to feel like NoA is Level 3, and it's important to squash this. That it's still very much an L2 system. The summary does a good job of that.

It's also important to make sure the driver monitoring requirements are understood for Level 3. The summary follow ups do a good job of that. This will likely be common as a lots of people will wonder why Tesla can't do Level 3.

 

[diplomat33]

That's a great job summarizing them.

Thank you.

I figured if anyone would try to summarize them it would be you, and to be honest I don't understand your obsession with them. They're great from an historical perspective on how we thought autonomous driving was going to go, but in reality they're not much use these days.

My "obsession" is based on my strong interest in autonomous driving in general and the fact that the levels are a classification system that is recognized and used by the autonomous driving industry and by regulatory agencies. So the SAE levels have real legitimacy. So it makes sense to try to understand them.

And I don't think the SAE levels are really trying to describe how autonomous driving will mature. The SAE levels are agnostic to the engineering problem of autonomous driving. That's because it's not the purpose of the SAE levels to tell companies HOW to do autonomous driving. Rather, the purpose is to describe different FUNCTIONS of autonomous driving systems so that we have a classification system that works regardless of how companies choose to do autonomous driving. For that, I think the levels work very well.

SAE Fantasy L1 -> Allow for either lane-steering or adaptive cruise control, but not both,.
Reality L1 -> Why would anyone do steering without adaptive cruise control? Oh, and the whole not requiring driving monitoring for adaptive cruise control didn't go so well with a society addicted to their cell phones. Any excuse to text is an excuse they'll use,.

SAE Fantasy L2 -> Allow for both lane-steering, and adaptive cruise control. Lets say it requires driver monitoring, but we're not going to say how or that it even has to be done well. We're also not going to put restrictions on how automated you can go while also still requiring human oversight even though it's been demonstrated time, and time again that humans can't oversee a highly automated task.
Reality L2 -> Lots and lots of police cars and fire trucks have been crashed into because of poor drive monitoring systems on L2 vehicles (especially with the Tesla emblem on them).

SAE Fantasy L3 -> Self-driving, but allow the manufacture to define the ODD. We won't define how well the driver monitoring has to be, and we'll completely forget about a persons tendency to fall asleep while being a passenger.
Reality L3 -> What Level 3? There is only one L3 car trying to get approval, but only in Germany. And, that car has some significant ODD limits to it's L3 capabilities. Which are allowed by the SAE Level 3, but seriously limit its usefulness.

SAE L4 -> We've done a lousy job defining three levels of automation so we should a better job on this one.
Reality L4 -> Pretty solid. Gives lots of room for improvements through expanding geo-fenced areas, and ODD capabilities.

SAE Fantasy L5 -> We're going to fantasize a lot on this one. Let's start by declaring all human drivers capable of the same ODD's, and make that the requirement.
Reality L5 -> Maybe next decade, maybe?

On a more serious note only two levels even apply to a new Tesla. Those levels are L2, and L4 where you if you don't buy FSD then you're stuck with L2. If you do buy FSD you might or might not eventually get an L4 experience with it.

L0, and L1 doesn't apply to Tesla because they all come with L2, and no one cares about L0 to L1 distinction.
L3 doesn't apply to Tesla because HW3 vehicles don't have the proper driver monitoring for that.
L5 doesn't apply to Tesla because HW3 can't handle the ODD situations that even the most average of average drivers can easily handle.

I know you are mocking the levels but that is not how the SAE came up with the levels. There is actually a very elegant logic to it.

SAE divides autonomous driving into 4 questions:
1) Who is controlling the x and y axis of motion of the car?
2) Who is responding to the changing environment around the car?
3) Who is responsible if the ADS has a problem?
4) When and where can the ADS work?

These are 4 important questions for autonomous driving.

The SAE came up with levels that are not a progression of features or capabilities but are progression of how many of these questions are answered by the ADS versus the human. The idea being that the ADS will be the answer for more of these questions as you go up the levels.

The progression is quite logical.

L0-L2 represent non autonomous driving where the ADS is not responsible for driving. What progression can there be when the ADS is not responsible for driving? It gradually controls more of the axis of motion without actually doing all the driving:

L0: manual driving
L1: System controls 1 axis of motion
L2: System controls 2 axis of motion

L3-L5 represent autonomous driving where the ADS is responsible for driving so now the ADS is responsible for question 1 and 2. It has to control both axis of motion and respond to the changing environment around the car.

L3: ADS answers questions 1 and 2 but driver answers question 3.
L4: ADS answers questions 1 and 2 and 3.

Then the SAE needed to answer question 4. At L4, the ADS is already doing everything, it's driving and taking care of failures. What's left? It made sense to have a L5 that would extend that capability to the broadest possible ODD. in other words, L3 and L4 say "we can do autonomous driving but only sometimes or in some areas". L5 says "I can do autonomous driving everywhere and anytime".

But, it's of very little actual information value especially to Tesla owners.

That's not the SAE's fault though. It's not the SAE's responsibility to come up with levels of autonomy that work for Tesla. The SAE's job is simply to describe autonomous driving in terms of general functionality. It's Tesla's job to develop an autonomous driving system for their cars as they see fit. And if Tesla is developing their autonomous driving system in a way that does not "work" for the levels, that's on Tesla.

 

[Daniel in SD]

Here is the latest I've read on the topic.
New study: Adaptive cruise-control and other driver ...https://www.seattletimes.com › seattle-news › transportation › new-study-ada...

I’m pretty sure any lane keeping system she combined with adaptive cruise control is a L2 system. That article says the vehicles were equipped with adaptive cruise and lane keeping.

But, it's of very little actual information value especially to Tesla owners.

It's valuable to me because it says what Tesla will need to do to make my car autonomous. Of course I didn't buy FSD because I don't think they can do it. Also it's valuable to me because I'm interested in autonomous vehicles in general, it seems like the most difficult engineering problem ever attempted.

The problem is the SAE levels leave a lot up to interpretation, and to regulators to define what's allowed and not allowed. That's why an L2 system in Europe is so much more constrained than an L2 systems in the US.

The problem is the safety of L2 systems is almost entirely a function of human factors. European culture is very different from American culture. There are many people here who seem to feel that even if Autopilot decreases overall safety it should still be allowed since when used correctly it provides personal benefit to the user.

The SAE Level 2 doesn't constrict Tesla on what they can do with it. They can implement the entire OEDR while still refusing to allow the car to be L4. The regulators on the other hand do care, and they will nerf it if they feel like if its unsafe or if its asking too much for the human to oversee it.

There is this:

The level of a driving automation system feature corresponds to the feature’s production design intent. This applies regardless of whether the vehicle on which it is equipped is a production vehicle already deployed in commerce, or a test vehicle that has yet to be deployed. As such, it is incorrect to classify a level 4 design-intended ADS feature equipped on a test vehicle as level 2 simply because on-road testing requires a test driver to supervise the feature while engaged, and to intervene if necessary to maintain safe operation.

Back in 2016 Uber tried to claim their autonomous vehicles were Level 2 since they required a test driver. The DMV decided that they were doing autonomous vehicle testing. I think that Tesla will not be allowed to operate software similar to what they showed on "autonomy day" as a Level 2 system.

Tesla could probably do driver monitoring for Level 3 with the driver facing camera. You only have to make sure the driver isn't sleeping and is in the front seat. Just use a neural net to filter out 99.9% of the cases that are easy and send a sampling of the rest to a data center in India. 1 month ban for the first offense, then 3 month, then... Just like a self driving car itself the driver monitoring system doesn't have to work 100% of the time.

 

[S4WRXTTCS]

SAE divides autonomous driving into 4 questions:
1) Who is controlling the x and y axis of motion of the car?
2) Who is responding to the changing environment around the car?
3) Who is responsible if the ADS has a problem?
4) When and where can the ADS work?

I think the SAE failed to ask the most important questions of them all. That question is what the human limitations are when it comes to their responsibility for Level 2 systems and Level 3 systems.

L0, and Level L1 are pretty straight forwards. Nothing about L1 induces the driver into inattention. I did argue that inattention happens, but its purposeful in the driver knows they are responsible but they send a text anyways. Nothing about adaptive cruise control or a ping ponging lane-keeping only safety system induces the person into inattention in an accidental way. Any inattention has to be more purposeful.

L2 should have been more restrictive in a lot of ways because of the inability for a human to understand the limitations of the systems, and the limitations of the humans ability to monitor not only the vehicle, but the driving as well. Basically the SAE Level 2 puts WAY too much on the driver. In the SAE's defense I don't think they really thought L2 systems would go beyond adaptive cruise control, and lane-steering.

L3 fails to take into account the humans ability to regain situational awareness. They only say "Several Seconds", but is that really enough time for a driver to take over?

It's Tesla's job to develop an autonomous driving system for their cars as they see fit. And if Tesla is developing their autonomous driving system in a way that does not "work" for the levels, that's on Tesla.

The problem with "that's on Tesla" is that Tesla distributed the industry, and that's pushing other car companies to follow Tesla lead. As an example currently Tesla is the only one with NoA as being Level 2, but I have no doubt other companies will do something similar and still list it as L2. But, going that far really goes beyond the scope of what the original authors of the SAE Levels really pictured in L2 system doing.

This is also a Tesla forum, and we both know Tesla completely ignored the SAE Levels. When referring to any Tesla implementation there isn't much point in bringing up SAE Levels. They didn't implement driver monitoring so even L2 systems don't work the way they were really supposed to.

I'm pretty convinced that the SAE authors would create a much different document now that they've seen what happened.

 

[S4WRXTTCS]

Back in 2016 Uber tried to claim their autonomous vehicles were Level 2 since they required a test driver. The DMV decided that they were doing autonomous vehicle testing. I think that Tesla will not be allowed to operate software similar to what they showed on "autonomy day" as a Level 2 system.

Tesla could probably do driver monitoring for Level 3 with the driver facing camera. You only have to make sure the driver isn't sleeping and is in the front seat. Just use a neural net to filter out 99.9% of the cases that are easy and send a sampling of the rest to a data center in India. 1 month ban for the first offense, then 3 month, then... Just like a self driving car itself the driver monitoring system doesn't have to work 100% of the time.

I completely agree with you on the first part, and I think we're going to see that this year. Where some regulatory body (likely the NHTSA) will restrict NoA.

On the second part I don't think a traditional CMOS camera can accomplish what a true driver monitoring system has. A true driver monitoring system can go much further in making sure the driver is alert. Plus I don't think Tesla is going to retrofit all the Model S/X vehicles to add it.

Instead I think the hammer is finally going to come down. Where the Level 2 driving will be limited by regulatory pressure in the US (like it currently is in Europe), and Tesla will have to prove it's capable of L4 driving. Hopefully it will be at least capable of geo restricted rest stop to rest stop travel or some supercharger to supercharger routes.

 

[S4WRXTTCS]

I’m pretty sure any lane keeping system she combined with adaptive cruise control is a L2 system. That article says the vehicles were equipped with adaptive cruise and lane keeping.

I'm pretty sure you're trying to get me to use the OP's summary to quote the line that tells you that you're wrong.

 

[diplomat33]

I think the SAE failed to ask the most important questions of them all. That question is what the human limitations are when it comes to their responsibility for Level 2 systems and Level 3 systems.

L0, and Level L1 are pretty straight forwards. Nothing about L1 induces the driver into inattention. I did argue that inattention happens, but its purposeful in the driver knows they are responsible but they send a text anyways. Nothing about adaptive cruise control or a ping ponging lane-keeping only safety system induces the person into inattention in an accidental way. Any inattention has to be more purposeful.

L2 should have been more restrictive in a lot of ways because of the inability for a human to understand the limitations of the systems, and the limitations of the humans ability to monitor not only the vehicle, but the driving as well. Basically the SAE Level 2 puts WAY too much on the driver. In the SAE's defense I don't think they really thought L2 systems would go beyond adaptive cruise control, and lane-steering.

You raise some good points that I don't necessarily disagree with. But your points may be a bit outside the scope of the SAE document which states it merely provides a "taxonomy with detailed definitions for six level of driving automation, ranging from no driving automation (level 0) to full driving automation (level 5), in the context of motor vehicles."

There are other documents such as the "safety first for automated driving" paper that definitely go into much more detail on the issues that you raise.

The problem with "that's on Tesla" is that Tesla distributed the industry, and that's pushing other car companies to follow Tesla lead. As an example currently Tesla is the only one with NoA as being Level 2, but I have no doubt other companies will do something similar and still list it as L2. But, going that far really goes beyond the scope of what the original authors of the SAE Levels really pictured in L2 system doing.

Yes, other companies are also developing NOA-like systems. Although, I would point out that they are all adopting more sophisticated driver monitoring systems, beyond what Tesla uses. Specifically, they use driver facing cameras that Tesla does not.

This is also a Tesla forum, and we both know Tesla completely ignored the SAE Levels. When referring to any Tesla implementation there isn't much point in bringing up SAE Levels. They didn't implement driver monitoring so even L2 systems don't work the way they were really supposed to.

I don't bring up the levels because of Tesla. I bring them up because I think they offer some value when discussing autonomous driving in general. Regardless of what Tesla does, we should still endeavor to be informed about autonomous driving in general, which involves understanding the SAE levels.

 

[Daniel in SD]

I'm pretty sure you're trying to get me to use the OP's summary to quote the line that tells you that you're wrong.

It’s longitudinal OR lateral control but not both for L1.

 

[S4WRXTTCS]

It’s longitudinal OR lateral control but not both for L1.

From the OP
"Active safety systems like automatic emergency braking or emergency lane departure avoidance can be included on cars of any levels."

Lane-Keeping is a Active Safety system, and not a convenience system like Lane-Steering.

The closest thing to Lane-Keeping that Tesla has is Lane Departure Avoidance, and that's an active safety system. Lane departure avoidance can either be configured to simply warn you or it can be configured to push you back into the lane similar to what a lane-keeping system does.

Lane-Keeping systems are often referred to ping-pong systems unlike lane-steering systems.