Seeing the world in autopilot, part deux

[bswn1]

Yes. This is nearing completion now. I am trying to get some interesting/diverse footage. There are going to be some limitations of course, and the backup camera is so unreliable I think I am not going to even try to depend on collect that particular footage.

I cannot wait to see what you two are putting together. Thanks from the community for your continued work and research.

 

[lunitiks]

Bladerskb, bladerskb, bladerskb!

 

[jimmy_d]

Well, in a stunning case of "ask and ye shall receive" Tesla obliged your request and added 3d bounding boxes in 18.39.6

(as always thnx to @DamianXVI for visualization and deciphering of data) (we are still debating how to bet visualize this)

View attachment 343923

Insane. You guys are awesome.

 

[strangecosmos]

Do the 3D bounding boxes entail a new system capability, or just a new visualization?

 

[lunitiks]

The visualization is @DamianXVI's work. Capability-wise it could be related to v9's "360 object decoder" that @jimmy_d talks about in his fantastic post over at the Neural Networks thread

 

[kbecks13]

(we are still debating how to bet visualize this)

WOW super cool! Could you give a quick example of what the data for these boxes looks like? Is it like box size + polar coordinates for box centroid + box orientation info?

 

[S4WRXTTCS]

stays in the slow lane too much

How German of it.

I finally get to try it out so I'm curious to see how my choices compare with its. I don't get to try the ULC yet, but I'll try to be a good obedient driver in changing lanes when it wants.

It does need to be regional because people drive differently in different regions. I fully expect it to be in the right lane more than me on a three lane road. I move over since I feel the middle lane is safer, and I only use the passing lane when I need to pass.

 

[jimmysanto]

Orange transparent “ribbon” denotes idea of how the autopilot thinks it should continue forward from there on (or so we think, it’s some sort of path planning apparently. It is given in 3D space and we @DamianXVI tried to overlaid it on 2D space as closely as could be done currently with the limited knowledge we have) – this is really advanced, note on the hills it goes up and down – this is why AP2+ is better on hilly roads than AP1.

@verygreen apart from the orange ribbon. I wonder if those lane lines are given in 3D space coordinate too?

 

[verygreen]

@verygreen apart from the orange ribbon. I wonder if those lane lines are given in 3D space coordinate too?

no 3D coordinates for lane lines.

 

[indiathebest]

@verygreen what's the output of aknet lane? coordinates on image space? or mathematical lane function?

 

[indiathebest]

@verygreen could you please reply it? i can't sleep well because of this..

 

[verygreen]

@verygreen could you please reply it? i can't sleep well because of this..

what do you mean by the aknat lane?

the lane lines are just a few points bezier curves in image space.

 

[indiathebest]

what do you mean by the aknat lane?

the lane lines are just a few points bezier curves in image space.

aknet lane means lane detection decoders in v9 neural network. so the network outputs bezier curve control points in image space. that's exactly what i've wondered. thank you very much!