How Zoox Uses Computer Vision To Advance Its Self-Driving Technology
HTML-код
- Опубликовано: 23 июл 2024
- Computer vision enables our autonomous vehicles to understand intricate details about their surroundings. For example, if a person is looking at their phone or whether flashing lights on a car mean it’s an emergency vehicle. Here's Sarah, Senior Director of Perception, explaining our approach - from instance and semantic segmentation, to classifying various pedestrian and vehicle behaviors.
zoox.com/careers/
#ComputerVision #ThisIsZoox #AutonomousVehicles #AutonomousMobility #SelfDriving 
Наука
Congrats and thank you ZOOX!!! so exciting the Future is here!!!
A feature I would like added is voice command override, in my country we have some ummm...security issues. There are instances where you don't actually want to avoid pedestrians you want to pass through them by force, yup you heard that right. Bad guys here can sometimes ask you to stop weapon in hand, with people driven vehicles you can take your chances by running through them, but with autonomous it will only require a gesture from a criminal to stop the car. The responsibility doesn't rest in the car or the company itself but in the person issuing the command of course. Anyway I don't want to sound grim just something that might need to be look upon in the future.
Way more impressive perception stack than I thought. I wonder if object permanence is created through all sensors or if cameras, lidar etc have their own
Really impressive classification stack, hope to see Zoox more often in my news feed)
Amazing articulation of Computer Vision
Wow, this is amazing guys!!
Cheers from RTI!
it gives me a lot of ideas for my computer vision projects :).
Learning self-driving car is like learn how to build a game in a real life. You guys are amazing.
This is the new future,❤️
Very cool! Great job with the skeleton! I never thought of that. But you know... context is kind of key. Like if you see a person in parked car that might help the system anticipate the car door is about to open.
Cheers from RemoTask Lidar Group
The most impressive thing to me about Zoox is how the driving agent is able to balance its value for safety with a value for forward progress. The confidence it demonstrates is incredible. I can't imagine a Tesla working through this same situation.
Well Tesla Vision or Full Self Driving is only Vision based and ultrasonic sensors , they don't use expensive and ugly Li-dars and radars like others . Also they are not Geo fenced like others.
It really looks awesome about taking that many classes of predictions and Detections.
But the hardware running these models would have been taking up a lot excited to know about it....
Amazing and fantastic.
I found this presentation of information very pleasant. Was she schooled in Glasgow?
We want Zoox to win. A great step forward for public transport.
God bless
amazing! and also there's a small dog/pet at 0:48 , it did't detect that might be less good features, hopefully it did't run over I guess. truck at 3:16 is quite dangerously detected, the wood log is left out, it could have come within box to be safe
Very good catch on both items. Never let anything go unnoticed. I may assume that the radar would pick up the overhang, but if the camera can pick up that as well and be incorporated into their sensor fusion model that would make for a better system. This is why so many miles have to be covered by autonomous vehicles in their testing phase. I can imagine companies such as Tesla, Zoox, etc., and how many exclaimed, " Didn't think of that!"
Could consider driver/pedestrian head orientation/gaze direction as well.
Zoox Come to New England states. After COVID. It looked like the test vehicle didn't see the car with the right turn signal changing lanes. You gotta have your sensors look for that. Can your sensors see flooded roads or mud slides or forest fires in its path and on the side of the roads?
Is the human pose estimation 2D or 3D
Me gusta este canal y esa empresa ♥️♥️🤤 pero no entiendo nada no traducen los vídeos al español 😞
How about the need of recognising people who are kissing, dancing, singing, art performing, fighting, arguing, carrying weapons, yelling, going on strikes and jumping on Roads? I hope the Zoox Software can cover those situations already. Am I right? It is interesting.
Do you detect someone who's on all four to pick up something on the road? Or a child picking up a toy that he dropped while he's in the crosswalk?
Инкредибл
To much abstraction I don't think this is going to work in new environment. I think reinforcement learning is a way forward for self driving car.