Cruise's rapid expansion announcement is starting to look like a threat.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
Tam
Well-Known Member
Don't machines deserve a mental day off sometimes?
That's what it seems like to SFMTA (the local transportation authority) too, but they don't have the power to directly ban the cars, all they can do is to protest to CPUC about the issuance of permits for driverless operation.
diplomat33
Average guy who loves autonomous vehicles
Well, if Cruise does not fix this issue and they try to scale to 1000's of cars, it could absolutely become a big problem.
Nice. That's what's needed for the real world. I bet they also detect and slow for dips and speedbumps.
diplomat33
Average guy who loves autonomous vehicles
Maybe. But if Cruise AVs cannot handle rain, then Cruise should keep a safety driver in them like Waymo has done. Waymo used safety drivers when it rained since they had not yet validated performance in all rain conditions yet. But now that Waymo knows that they can handle rain, wet roads and puddles safely and smoothly, they will likely remove safety drivers. But they kept safety drivers before when they were not sure yet.
diplomat33
Average guy who loves autonomous vehicles
diplomat33
Average guy who loves autonomous vehicles
Ghost Autonomy just got a permit from CPUC to allow the public to ride in their autonomous cars. They are starting with highway autonomous driving for consumer cars because they believe it is the fastest path to transforming daily driving.
www.ghostautonomy.com
I had not heard of Ghost Autonomy before but apparently they are a software company that is focused on developing the AI for self-driving. Their mission is to bring self-driving to everyone by building self-driving software that can work on any average consumer car.
They say they have a breakthrough NN that can generalize and understand the physics of obstacles in motion allowing the car to avoid collisions even in the long tail of edge cases:
They are starting with highway driving but plan to eventually scale to other road types:
Here is their description page:
www.ghostautonomy.com
Bringing Consumer Autonomy to California’s Highways | Ghost
Ghost is first to secure autonomous driving permit for consumer cars
www.ghostautonomy.com
I had not heard of Ghost Autonomy before but apparently they are a software company that is focused on developing the AI for self-driving. Their mission is to bring self-driving to everyone by building self-driving software that can work on any average consumer car.
They say they have a breakthrough NN that can generalize and understand the physics of obstacles in motion allowing the car to avoid collisions even in the long tail of edge cases:
They are starting with highway driving but plan to eventually scale to other road types:
Here is their description page:
Self-Driving For Everyone | Ghost
A breakthrough in self-driving, built for safety, scale, and ultimately freedom.
www.ghostautonomy.com
Last edited:
diplomat33
Average guy who loves autonomous vehicles
Here is an interesting video on Ghost Autonomy's physics based AI approach:
Basically, their approach is to develop AI that can detect objects in general based on shape and motion without explicitly training it to classify the object first. Put simply, they train the NN to recognize objects based on their general shape and how they move instead of training the NN to classify specific images.
They believe this is a more generalized approach because it will work on any object, even if the AI does not know what the object is. So it should work on all edge cases, even ones that you did not explicitly train for. They believe this is better than traditional computer vision that relies on object classification because it avoids that long tail problem of edge cases you have not trained on yet.
Here is the video where they discuss why they believe traditional computer vision cannot solve the long tail:
Basically, their approach is to develop AI that can detect objects in general based on shape and motion without explicitly training it to classify the object first. Put simply, they train the NN to recognize objects based on their general shape and how they move instead of training the NN to classify specific images.
They believe this is a more generalized approach because it will work on any object, even if the AI does not know what the object is. So it should work on all edge cases, even ones that you did not explicitly train for. They believe this is better than traditional computer vision that relies on object classification because it avoids that long tail problem of edge cases you have not trained on yet.
Here is the video where they discuss why they believe traditional computer vision cannot solve the long tail:
Last edited:
Ghost Autonomy just got a permit from CPUC to allow the public to ride in their autonomous cars. They are starting with highway autonomous driving for consumer cars because they believe it is the fastest path to transforming daily driving.
Bringing Consumer Autonomy to California’s Highways | Ghost
Ghost is first to secure autonomous driving permit for consumer cars
I had not heard of Ghost Autonomy before but apparently they are a software company that is focused on developing the AI for self-driving. Their mission is to bring self-driving to everyone by building self-driving software that can work on any average consumer car.
They say they have a breakthrough NN that can generalize and understand the physics of obstacles in motion allowing the car to avoid collisions even in the long tail of edge cases:
They are starting with highway driving but plan to eventually scale to other road types:
Here is their description page:
Self-Driving For Everyone | Ghost
A breakthrough in self-driving, built for safety, scale, and ultimately freedom.
Interesting. I like their use of multiple sensors and goal of hands free driving. It looks like a small garage shop with a long road ahead. One online source says they have have a sizeable $157M debt owed to investors. They'll have to sell systems for well over $2k to break even. At least they are doing their homework up front.
diplomat33
Average guy who loves autonomous vehicles
They use 48 MP cameras and imaging radar.
They are obviously a very small start-up hoping their "physics based AI" will allow them to develop a cheap and generalized self-driving that they can license to automakers. The fact that they got a permit to let the public ride in their autonomous cars is encouraging that they have some sort of prototype. And I do like that they are focused on highway driving first. If their approach works, they could potentially deliver L4 highway to lots of consumer cars. That sounds like a great business model to me. Heck, I would happily pay $2k for attention-free highway driving. I don't think they are a serious threat to Waymo or even Tesla. But if they can deliver attention-free highway driving on consumer cars that would be a great product IMO.
Here is a quick video of their "attention free highway driving" in action.
And another video that explains more about their tech and approach:
Last edited:
That's a very similar idea to the Occupancy Network Tesla is working on. The previous NN did depend on classifying specific images and that did make it so objects it didn't recognize, it wouldn't properly respond to. The ON is a lot more efficiency because it only recognizes occupied space, it doesn't have to recognize what object is there (although labels are optionally added to certain blocks).
A Look at Tesla's Occupancy Networks
diplomat33
Average guy who loves autonomous vehicles
Both approaches share some similarities. They are both attempts to get around the problems with classic object classification. But I think Ghost's approach is very different from occupancy networks. Occupancy networks just look at if a space is occupied or not by something you don't want to hit. Ghost is detecting objects based on their shapes and motion.
diplomat33
Average guy who loves autonomous vehicles
Zoox CEO believes autonomous driving is "one generation away from achieving critical mass" with slow and steady progress and consolidation in the mean time:
EVNow
Well-Known Member
If they could bring autonomy to 15 to 25 year old cars that would be amazing
We are always a few years away from AVs that can take you anywhere at anytime Traveling on any road in any type of Weather
The Ghost speaker makes it sound as if all other systems suffer from the inability to see and react to unlabelled or unclassified objects. I would say straight away that that cannot be true. It is utterly obvious nonsense.
They want to tell us that, for example, a Tesla completely ignores an object in the street that it cannot categorize? Nobody in his right mind would program a car like that. Of course an automated car will be programmed to stop in front of an uncategorized object, rather than crash into it.
This makes the whole interview rather questionable. I won't bet on Ghost.
They want to tell us that, for example, a Tesla completely ignores an object in the street that it cannot categorize? Nobody in his right mind would program a car like that. Of course an automated car will be programmed to stop in front of an uncategorized object, rather than crash into it.
This makes the whole interview rather questionable. I won't bet on Ghost.
Similar threads
- Article
