Tim Y.C 🌐🌍Self driving cars is a cultural phenomenon that could only arise on the west coast USA. No sane individual would look at the streets of NYC and convince themselves that they could get a car to navigate through it on its own. Are there even any self driving companies based out of the East Coast USA? @notjustbikes #autopilot #selfDrivingCar #tesla #publictransit #transit
michelle.cat@clsytim @notjustbikes Drones need special FAA clearance to fly in flight restricted areas something similar can be done with autonomous cars the geoboundaries change periodically based on a number of conditions open highways on a clear day yes , highways in inclement weather no , busy surface roads no
@clsytim @notjustbikes commercial services might get more latitude likewise different restrictions if there is a driver on board or not
@clsytim @notjustbikes well it looks like Tesla is admitting defeat it was an ignorant move removing radar , sacrificing safety over costs probably too late to avoid lawsuits though https://electrek.co/2022/12/06/tesla-radar-car-next-month-self-driving-suite-concerns/
@clsytim @notjustbikes This is the most ignorant argument for camera only systems I have heard “The probability of safety will be higher with pure vision than vision+radar, not lower. Vision has become so good that radar actually reduces signal/noise.” It’s not like these systems work sequentially and additional sensors are more inputs for the neural net , these are asynchronous systems that trigger alerts , a radar event should have higher polling priority
@clsytim @notjustbikes if LiDAR or radar detects and obstacle it should trigger a braking event or return driver control regardless if the computer vision system detected it or. Not , take a space launch a single negative on a polled system can cancel as launch more sensors in self driving cars make them safer not less safer
@mcvresearch personally im bearish on sensor fusion system that depends on vision as primary input. While I understand the argument is "humans use vision...so...duh", but we actually use vision for obstacle avoidance mainly in the same way as a point cloud system that a LIDAR produces. We recognize distances to edges and vertices and make decisions based on those. You can actually see or reliance on this very easily when you put a human in a "mirror house". We still have "vision" but it is useless without understanding bounds on edges and vertices.
You could further argue that you can develop these point clouds from stereoscopic vision, but, one, you have a lot of noise and, two, i dont think you will be able to compute results at the same fortitude as millions of years of evolution has ingrained in humans #selfDrivingCar #computer #computervision #AI
@clsytim agree you see by my posts I am not an advocate of vision only systems
@mcvresearch but also, disclaimer, i could totally be out of my league here. Its been roughly 3 years since I did anything advanced in computer vision and im sure the field has progressed since then
CurtAdams@mcvresearch @clsytim @notjustbikes Tesla's story about cameras only being better was a transparent cover up. In spring 2021 when the chip shortage was at its worst Tesla couldn't ship cars because it couldn't get chips for the radar. So it said it didn't need them and shipped them anyway. Always a ridiculous claim. Current Teslas can't even parallel park.
@CurtAdams @mcvresearch @notjustbikes yeah, Ive heard this speculation before. Also, tbf, I have been in a Tesla that has parallel parked. Doesnt nail it 100%, but when it works its pretty cool.
Bro@bro @clsytim @CurtAdams @notjustbikes I am sure Teslas are decent cars as long as they have a human driver , I have nothing against them except for a concussion
Watch: Tesla In Autopilot Crashes Into North Carolina Police Vehicle
@clsytim @mcvresearch @notjustbikes The old Teslas can parallel park but the new ones can't. My brother in law just got a new model Y and is passed about it not being able to.
@mcvresearch @notjustbikes precise geobounding is actually quite tricky. It's even harder in cities with skyscrapers where GPS get scattered easily. I know uber was doing research to reduce error from this but it was mostly mitigation not a full resolution.
@clsytim @notjustbikes I don’t think they need to be precise though LANNC zones for instance are pretty broad when it comes to restricted zones such as airports and hospitals
@clsytim @notjustbikes and note you also take weather conditions into account whereas full autonomous mode might be fine on an interstate in clear weather but be restricted during a snow storm