Assessment of Operational Design Domain for Level 2 autonomous vehicles.

[Shubham]

Hello everyone,

My name is Shubham Bhusari and I am a MSc student in TU Delft, Netherlands.
I am currently working on a project in order to understand the conditions and situations in which the autopilot function of Tesla fails along with understanding the drivers trust in the system around edge cases. My objective is to formulate a standard definition of Operational Design Domain for level 2 Automated vehicles across all manufacturers. I aim to do so by first understanding the drivers needs and trust in edge cases through controlled on-road tests and then combine it with objective risk measurements using various scientific methodologies. I focus on Tesla, because it has been in the market for more time than the other manufacturers and from prior studies here at the university, it is seen that they perform the best across various situations. Since safety is of utmost importance in our studies, Tesla was chosen.

With this thread, I wanted to know if anyone of you Tesla owners know a few situations especially on highways where the system performance is ambiguous and thereby affects your trust in the system. Or maybe even specific routes in the Netherlands which could be a candidate route for my road tests?

If yes, Please feel free to mention a few such situations here in this thread, to have a small discussion about them as this would be really helpful for my project.

Looking forward to hear from all of you, as the feedback of experienced drivers like you is very important.

P.S-I have nothing against Tesla or as a matter of fact any vehicle manufacturer and my ultimate aim is to ensure that manufacturers better inform their users which in-turn leads to more road safety over time.

 

[BobbyKings]

There is a lot of good videos on this thread that may help you:

Software Update 2018.10.4

Nothing beats TU Delft

 

[basvk]

Concerning road tests: roads that have recently changed or roads that are not digitized correctly into maps are guaranteed to give you interesting behaviour with autopilot.

For me the biggest ambiguity in (semi) autonomous driving or even TACC is how the system handles non-moving cars and objects (it often ignores them) and moving cars and/or objects (they are identified and avoided). So I'm always wondering whether the car will brake in time when approaching slow traffic or road works for example, which kind of defeats the whole autonomous promise.

 

[BobbyKings]

Some more info here:

Latest Autopilot Update Gets Cheers From Tesla Owners | CleanTechnica

 

[BobbyKings]

En hier:

 

[BobbyKings]

Verrazano u-turn:

 

[Wim S.]

Verrazano u-turn:

Die jongeman wil helemaal niet weten hoe goed de AutoPilot is. Hij wil weten wat wij er niet goed aan vinden

 

[BobbyKings]

En met sneeuw:

 

[BobbyKings]

Die jongeman wil helemaal niet weten hoe goed de AutoPilot is. Hij wil weten wat wij er niet goed aan vinden

Ah OK. Dan is het nu even genoeg.

 

[BobbyKings]

Nou ja deze dan nog.

 

[BobbyKings]

Toch mag deze uitgebreide video van Björn met slecht licht en slechte lijnen ook niet ontbreken.

Nu klaar.

 

[is3]

En sluiten maar weer dit draadje........

 

[Shubham]

There is a lot of good videos on this thread that may help you:

Software Update 2018.10.4

Nothing beats TU Delft

Thanks a lot for your help bobbykings

 

[Shubham]

Concerning road tests: roads that have recently changed or roads that are not digitized correctly into maps are guaranteed to give you interesting behaviour with autopilot.

For me the biggest ambiguity in (semi) autonomous driving or even TACC is how the system handles non-moving cars and objects (it often ignores them) and moving cars and/or objects (they are identified and avoided). So I'm always wondering whether the car will brake in time when approaching slow traffic or road works for example, which kind of defeats the whole autonomous promise.

thanks for your valuable feedback behavior around stationary or slow moving vehicles is indeed a factor I will consider for my project.

 

[Shubham]

Hello everyone, thanks for your feedbacks and comments so far, please keep them coming.
Furthermore, I would also like to know if there any specific roads in the netherlands that anyone finds it difficult to use the AP? This will help in identifying the candidate routes for my on-road test and thereby help me in understanding what is going wrong at those locations to ultimately recommend solutions to avoid them.

Looking forward to your comments

Cheers!
Shubham Bhusari

 

[pvkeep]

Personally, I'm pretty confident when using AP(1) 95% of the time. There are a few situations where I am more alert to possible AP problems when driving on motorways. These are:
- A44 bridges, there are three of them on that road, all constitute a possible AP behaviour anomaly.
- Bridge underpasses where the road dips a couple of meters. Leiden has a few of these.
- Sharp turns in the road, lots of examples there
- Hilly environments where the AP camera and radar have limited view of what happens beyond the next hill
The last one would only apply to Limburg in the Netherlands, my experience there is with secundary roads in France.

 

[SuperDave]

Hello everyone, thanks for your feedbacks and comments so far, please keep them coming.
Furthermore, I would also like to know if there any specific roads in the netherlands that anyone finds it difficult to use the AP? This will help in identifying the candidate routes for my on-road test and thereby help me in understanding what is going wrong at those locations to ultimately recommend solutions to avoid them.

Looking forward to your comments

Cheers!
Shubham Bhusari

A1 nieuwe weg bij Amsterdam

 

[pvkeep]

Last night on the A1 I had another situation where my trust in AP was temporarily reduced: a testicle-bumblebee was checking his phone instead of concentrating on the road. We as humans can detect pretty quickly that another car is showing erratic driving behaviour, these days almost always caused by the driver reading WhatsApp or whatever. This puts me on high alert, ready to take evasive manoeuvres if the other car drifts into my lane whilst I'm overtaking it. AP has no such concept and will happily stay in the centre of the current lane.

 

[Shubham]

Personally, I'm pretty confident when using AP(1) 95% of the time. There are a few situations where I am more alert to possible AP problems when driving on motorways. These are:
- A44 bridges, there are three of them on that road, all constitute a possible AP behaviour anomaly.
- Bridge underpasses where the road dips a couple of meters. Leiden has a few of these.
- Sharp turns in the road, lots of examples there
- Hilly environments where the AP camera and radar have limited view of what happens beyond the next hill
The last one would only apply to Limburg in the Netherlands, my experience there is with secundary roads in France.

Hi Paul, thanks for your comments. I will certainly look through these routes and identify what type of anomalies exist there.

 

[Shubham]

Last night on the A1 I had another situation where my trust in AP was temporarily reduced: a testicle-bumblebee was checking his phone instead of concentrating on the road. We as humans can detect pretty quickly that another car is showing erratic driving behaviour, these days almost always caused by the driver reading WhatsApp or whatever. This puts me on high alert, ready to take evasive manoeuvres if the other car drifts into my lane whilst I'm overtaking it. AP has no such concept and will happily stay in the centre of the current lane.

A1 nieuwe weg bij Amsterdam

Hi Dave, thanks for your inputs. A1 is certainly in my list of candidate routes for my on-road test.