Discuss: Waymo Explains Their Multimodal Approach to Autonomy [VIDEO]

[Not a Tesla App]

As Tesla's vision-only autonomy architecture gains ground, Waymo continues to advocate for a multimodal sensor suite combining cameras, radar, and LiDAR Read More

Waymo Explains Their Multimodal Approach to Autonomy [VIDEO] - Not a Tesla App

As Tesla's vision-only autonomy architecture gains ground, Waymo continues to advocate for a multimodal sensor suite combining cameras, radar, and LiDAR

www.notateslaapp.com

 

[ALIEN SUPERSTAR]

this was very interesting! i’m in london and i saw a waymo and my jaw did drop at first because i had no idea they were testing in the UK.

 

[Procal]

I don't mind redundancy, especially with a system where that is the expectation. My issue is that now having taken a ride in both a Waymo and a Tesla robotaxi in Miami, the robotaxi is significantly more humanlike and smooth than the Waymo. If I had to describe it, the Waymo felt almost similar to how early versions of v12 or v13 FSD felt, with a ton of jitters, sudden steering wheel jerks, tapping of the brakes, etc.

What I feel will become increasingly more clear is that while additional sensors help with the perception and redundancy aspect, the software that actually controls the car and decides how to use that information will be equally if not significantly more important. If you have great sensors but the car doesn't know how to react in complex scenarios, how to handle edge cases appropriately, or feel smooth, then you just have good sensors with a dumb brain.

My other slightly less serious concern would be the additional energy required for the compute and additional sensors that would be needed. However, that is less of a factor when safety is the highest priority.

 

[TESLASUPERFAN]

Older teslas 2024 and under still have USS , they could use this in some aspect as a redundancy. Moving forward they could easily retrofit cars with new USS…….or am I missing something ?

 

[Procal]

Older teslas 2024 and under still have USS , they could use this in some aspect as a redundancy. Moving forward they could easily retrofit cars with new USS…….or am I missing something ?

USS is typically very short range; I don't know how useful that would be outside of very slow-speed maneuvers, for example, a parking lot. LiDAR and radar, by extension, have more usefulness in more areas, but both have different qualities in resolution and the environment they can "draw." For example, a radar might be able to tell you that there is "something" in front of the car, but it won't be able to make out what it is. A LiDAR, on the other hand, would be able to tell you the exact shape, size, and distance to that "thing." But all of this feeds back to the "brain," which is what really matters, because that is what ultimately decides what happens with that information. It's basically just more ways of "seeing" the world, and each one overlaps with the shortcomings of the other.

 

[TESLASUPERFAN]

dont most fsd fails happen at relatively low speeds, like within the city. I would think tesla vision collaborating with new redesigned USS could work for say parking (banish/ summon).

Id remain skeptical til I see fsd 15 in an AI5 vehicle. Theres a new kid on the block for self driving cars called WAVY came out this week with safety drivers, looks promising but they are brand new so we will have to see.

 

[Procal]

dont most fsd fails happen at relatively low speeds, like within the city. I would think tesla vision collaborating with new redesigned USS could work for say parking (banish/ summon).

Id remain skeptical til I see fsd 15 in an AI5 vehicle. Theres a new kid on the block for self driving cars called WAVY came out this week with safety drivers, looks promising but they are brand new so we will have to see.

I wouldn't say fail necessarily. Most FSD disengagements on HW4 at low speeds, I assume, are in parking lots and are mostly due to parking preferences and things of that nature, not because they are hitting things or would hit things. Nonetheless, I believe that Tesla is currently trying to "solve autonomy" first and then figure out what is actually required to make it work better. That means if they see that they really can't get any farther with just cameras alone, then they may end up adding more sensors in the future. We've already seen them doing something similar with adding an additional camera to the front bumper of newer cars and camera washers for the Cybercab. I think they are just waiting until they hit that bottleneck rather than trying to add things that they may not end up needing or removing later, which I think is the wrong approach, but I guess time will tell.