References: ►Read the full article: www.louisbouchard.ai/waymo-lidar/ ►Piergiovanni, A.J., Casser, V., Ryoo, M.S. and Angelova, A., 2021. 4d-net for learned multi-modal alignment. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 15435-15445). ►Google Research's blog post: ai.googleblog.com/2022/02/4d-net-learning-multi-modal-alignment.html?m=1 ►My Newsletter (A new AI application explained weekly to your emails!): www.louisbouchard.ai/newsletter/
I would say the future of self-driving cars would lie in such a way that less sensors but more cameras are built into the car to almost build a world like simulation that is as much close to the human eye view as possible and able to co-ordinate the ride with safety features implemented in real-time. So more or less safety is important and in real world it would reduce the risk of failures. But the concept is really one of those subsets of AI that is coming closer to practical use faster than ever.
Indeed! I’m excited to see what will happen. What do you think will end up winning? Tesla’s “humanoid” way of using only cameras, or other companies approach of using as many useful sensors as possible? :)
@@WhatsAI tesla is going to win, not because just cameras are better but because not using the rest makes it scalable, and scalability is the issue with teaching AI drivers, not quality/quantity of information the AI is receiving while driving. In theory cars shouldn't have worked, anyone can feed a horse, you don't need to study engineering and understand engine mechanics to get more horses, they can go basically anywhere, maintenance is much lower. It wasn't a company that built the best, most advanced car that lead to seeing cars everywhere in the world, it was the one that made it simple and affordable. The advancements that we see today when comparing the most basic models to the modelT came slowly in the century that followed. Few/close to nobody is using AI drivers today because they aren't robust enough for general usage or cheap enough to scale it up yet, real life isn't a sci-fi movie where you can just do a montaje after someone builds a prototype of future tech to everyone in the world using it. It's a little ironic how companies like Google still dump money into things like this when you have Tesla doing the opposite, it's like they still haven't leaned anything from their now famous google graveyard where most things died mainly because they were out of touch, either too complicated or expensive for normal people to adopt.
I completely agree with you. These are indeed strong arguments in favour of Tesla! I’d just add that we are also not using AI drivers yet because they are not legally allowed haha, at least in most countries/states. But I completely agree with you!
@@WhatsAI yea, that's an important reason too. But I personally tend to place it in the same umbrella of people's fear of complicated things breaking, I'm sure it was a prevalent fear for those buying cars one century ago, and I could see it just little over a decade ago when I was trying to get my parents to start using the tablet phones over the old Nokia button phones, I remember their grievances were that it seemed complicated and they wanted something simple and were afraid to mess up, now they use them everyday like it's nothing. I'm sure something similar happened to many others gaining trust to adopt new tech because they see a lot of people around them also using it without problems. I think the same is/will happen with AI drivers. That's why I tend to consider the traffic laws issue just another part of the adoption issue.
@@WhatsAI Yeah anytime! Do you have any advice or recommendations for people starting out new in CS? College, Bootcamp or self taught? I am currently self teaching and find it a bit hard to understand some topics. Is college or code bootcamp worth it? Thanks
I’ve done the traditional college path and I can say that, even doing this, you need to learn on your own online. So I’d say self taught is definitely needed in all cases! Still, I’d recommend getting a degree if you can. It still is necessary for most bigger companies unless you create amazing products or research on your own and build a strong portfolio! I believe college is worth it if you really love the theory and research, especially for ML, but not worth at all for CS. I’d rather focus on creating online products and applications useful for people and learn doing it. You can take some courses as a free student that you find interesting if you wish to compliment a specific topic, it certainly won’t hurt you!
@@WhatsAI Thanks for the recommendations, I really appreciate it. Do you know where I can find free courses? Been looking online but seem to run into scam sites or bootcamp promotions.
For courses I’d recommend investing in good ones, so paid ones. But there are amazing free courses on youtube as well, as shared in my guide for starting in ML! You can look for the stanford ones where you can also find home-works and answers to practice!
I agree! But it also means more information which can be discarded if in failure mode or giving wrong information with great algorithms. I’d say it really depends on the tech? Both on great sensors and great curation of the information captured by those sensors
@@WhatsAI Agree in the end discarded info is also data to help improve technology, but I’m intrigued in the standards that will be used to measure this techs specially with human lives involved
People simplify their driving by the use of "assumptions." They, using experience assume that which they see in front of them is a "road." They assume, by way of simple rules (red light, green light, lane, other traffic etc) that the road is clear, while looking for exceptions. So far this isn't happening with AI for vehicles.
References:
►Read the full article: www.louisbouchard.ai/waymo-lidar/
►Piergiovanni, A.J., Casser, V., Ryoo, M.S. and Angelova, A., 2021. 4d-net for learned multi-modal alignment. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 15435-15445).
►Google Research's blog post: ai.googleblog.com/2022/02/4d-net-learning-multi-modal-alignment.html?m=1
►My Newsletter (A new AI application explained weekly to your emails!): www.louisbouchard.ai/newsletter/
I would say the future of self-driving cars would lie in such a way that less sensors but more cameras are built into the car to almost build a world like simulation that is as much close to the human eye view as possible and able to co-ordinate the ride with safety features implemented in real-time. So more or less safety is important and in real world it would reduce the risk of failures. But the concept is really one of those subsets of AI that is coming closer to practical use faster than ever.
I am curious who will make the race. Tesla with using only RGB cameras or waymo with LiDAR+cameras
Indeed! I’m excited to see what will happen. What do you think will end up winning? Tesla’s “humanoid” way of using only cameras, or other companies approach of using as many useful sensors as possible? :)
@@WhatsAI tesla is going to win, not because just cameras are better but because not using the rest makes it scalable, and scalability is the issue with teaching AI drivers, not quality/quantity of information the AI is receiving while driving.
In theory cars shouldn't have worked, anyone can feed a horse, you don't need to study engineering and understand engine mechanics to get more horses, they can go basically anywhere, maintenance is much lower. It wasn't a company that built the best, most advanced car that lead to seeing cars everywhere in the world, it was the one that made it simple and affordable. The advancements that we see today when comparing the most basic models to the modelT came slowly in the century that followed.
Few/close to nobody is using AI drivers today because they aren't robust enough for general usage or cheap enough to scale it up yet, real life isn't a sci-fi movie where you can just do a montaje after someone builds a prototype of future tech to everyone in the world using it.
It's a little ironic how companies like Google still dump money into things like this when you have Tesla doing the opposite, it's like they still haven't leaned anything from their now famous google graveyard where most things died mainly because they were out of touch, either too complicated or expensive for normal people to adopt.
I completely agree with you. These are indeed strong arguments in favour of Tesla! I’d just add that we are also not using AI drivers yet because they are not legally allowed haha, at least in most countries/states. But I completely agree with you!
@@Kmykzy well put. scalability is key
@@WhatsAI yea, that's an important reason too. But I personally tend to place it in the same umbrella of people's fear of complicated things breaking, I'm sure it was a prevalent fear for those buying cars one century ago, and I could see it just little over a decade ago when I was trying to get my parents to start using the tablet phones over the old Nokia button phones, I remember their grievances were that it seemed complicated and they wanted something simple and were afraid to mess up, now they use them everyday like it's nothing. I'm sure something similar happened to many others gaining trust to adopt new tech because they see a lot of people around them also using it without problems. I think the same is/will happen with AI drivers. That's why I tend to consider the traffic laws issue just another part of the adoption issue.
Great information! Keep up the great content 👍
Thank you very much!! 🙏😊
@@WhatsAI Yeah anytime! Do you have any advice or recommendations for people starting out new in CS? College, Bootcamp or self taught? I am currently self teaching and find it a bit hard to understand some topics. Is college or code bootcamp worth it?
Thanks
I’ve done the traditional college path and I can say that, even doing this, you need to learn on your own online. So I’d say self taught is definitely needed in all cases! Still, I’d recommend getting a degree if you can. It still is necessary for most bigger companies unless you create amazing products or research on your own and build a strong portfolio! I believe college is worth it if you really love the theory and research, especially for ML, but not worth at all for CS. I’d rather focus on creating online products and applications useful for people and learn doing it. You can take some courses as a free student that you find interesting if you wish to compliment a specific topic, it certainly won’t hurt you!
@@WhatsAI Thanks for the recommendations, I really appreciate it.
Do you know where I can find free courses? Been looking online but seem to run into scam sites or bootcamp promotions.
For courses I’d recommend investing in good ones, so paid ones. But there are amazing free courses on youtube as well, as shared in my guide for starting in ML! You can look for the stanford ones where you can also find home-works and answers to practice!
I’m thinking more sensors means more points of fail. The regulations of this technologies will be even more brutal as this tech move forward
I agree! But it also means more information which can be discarded if in failure mode or giving wrong information with great algorithms. I’d say it really depends on the tech? Both on great sensors and great curation of the information captured by those sensors
@@WhatsAI Agree in the end discarded info is also data to help improve technology, but I’m intrigued in the standards that will be used to measure this techs specially with human lives involved
People simplify their driving by the use of "assumptions." They, using experience assume that which they see in front of them is a "road." They assume, by way of simple rules (red light, green light, lane, other traffic etc) that the road is clear, while looking for exceptions.
So far this isn't happening with AI for vehicles.
So umbrellas fool LiDAR... got it! Better wet then hit.
very nice
Great content ..... as always
Thank you very much! 😊
Lidar is soo cool! Maping out the real world with lasers, radars, point clouds and deep learning!
nicely explained..thanks for the detailed video :)
Thank you so much Abhishek! 🙏