Sandy's Trip to Ann Arbor | Automated Vehicle Education at Mcity

Sandy and the Munro Live Team take a trip to the University of Michigan to check out the latest research on automated vehicles and how they are testing them under controlled but realistic conditions.
If you like the content you're seeing, please consider visiting our website or subscribing to our Patreon group for exclusive content.
PATREON GROUP: / munrolive
MUNRO LIVE LINKEDIN: / munro-live
MUNRO MERCH: teespring.com/stores/munro-li...
MUNRO LIVE WEBSITE: munrolive.com/
MUNRO LIVE TWITTER: / live_munro
MUNRO LIVE INSTAGRAM: / munrolive
MUNRO LIVE FACEBOOK: / munroliveteam
MUNROLIVE TIK TOK: / munrolive
MUNRO & ASSOCIATES MAIN WEBSITE: leandesign.com/
Outro:
Music from KZbin Audio Library [Aka YAL]
Music provided by RFM: • Video
#SandyMunro #autonomousvehicles #UniversityofMichigan

[@bjs2022]

Thank you for the good, professional sound. (Everybody has their own body microphone.)

[@fred993a]

Sandy - you are appreciated so much. Thank you.

[@syweiner434]

I intuitively agree that smart-roads could greatly accelerate FSD implementation but: 1. realistically, the funding, development and rollout of STANDARDIZED "smart-road" technology will likely take decades. 2. centralized control of all the vehicles by the gubermint (ala CCCP) holds many implications for us all. 3. each car needs FSD capability anyway, to prevent the "mother of all traffic jams" in case the smart-roads fail (or are hacked).

[@davidrowewtl6811]

Their idealism vs Tesla's realism.

[@7sq]

Yes

[@b4804514]

Communication should be required between vehicles IMO

[@davidrowewtl6811]

@@b4804514 It sort of is already, isn't it? Be it you, I or Tesla AI we see cars, we plan their expected position and adjust ours, or not, accordingly. All the time they are doing the same for/with us.

[@toddmarshall7573]

15:45 "...I don't see a lot of proposals from city councils...": Earth to Sandy. If you choose a government solution to any problem, you're still looking for a solution.

[@danielhull9079]

Or Blackberry QNX [V2X]?

[@Weezedog]

That’s a traffic light tree art sculpture in London, England. It’s not used to direct traffic, it’s just an art piece.

[@jamesbell6182]

Tell your self driver that not a traffic controller and see how that works

[@budgetaudiophilelife-long5461]

😂🤗 THANKS FOR SHARING THIS…A BIG RELIEF 😅

[@deltajohnny]

It seems that Sandy thought that it was real 😉😉

[@budgetaudiophilelife-long5461]

@@deltajohnny 👍 I’m afraid I did two until I read it in the comments it was art

[@deltajohnny]

@@budgetaudiophilelife-long5461 I saw this modified picture some time ago in Internet, but if I had not seen it before, it seemed real! 😉😉

[@dogtato]

it's undoubtedly easier to train a car to handle bad roads than it is to fix all the roads. one is at least possible in theory. i think if you pushed someone in that modified wheelchair with the bars in the back, you would soon need your own, but maybe the design doesn't murder shins like it looks it would, or it's a proof of concept just for testing crashes

[@terrysullivan1992]

Had a wheelchair van for my mom 18 years ago. the docking system was similar to a 5th wheel where the chair had the vertical post and the docking plate was mounted to the floor of the van. It was straight in from the back into the passenger position facing forward. Then there was an adapter to use the existing seat belts. Worked very well, but probably would not have functioned all that well in a 30 mpg crash. Nonetheless; it made a huge difference for my mothers mobility.

[@MrFoxRobert]

Thank you!

[@MunroLive]

You're welcome!

[@gbinman]

I have thought for a long time that many issues could be resolved at the infrastructure level more efficiently than making super smart vehicle systems. The acceptance issues as illustrated in the horse vs car scenario is exactly where we are at with this transition. There are many forces that want this to fail.

[@aurelio-reymilaorcabal9669]

This is where China holds an edge, they can dictate and put up the infrastructure in that country plus they have Huawei and 5G network, they already have some lev3-4 equipped cars on the road , i think Xpeng.

[@CHIEF_420]

🎓

[@ruftime]

Thank you! How about adopting an AIS type system for all, or just emergency vehicles with running lights? Automatic Identification System (AIS) is a short-range coastal tracking system currently used on ships.

[@TFB-chris]

The traffic light tree is just a sculpture.

[@terrysullivan1992]

AIS for all vehicles to warn of emergency vehicle traffic would be good. I notice lots of drivers don't hear these approaching and it is often difficult to figure out where they are and where they are headed.

[@user-to2rf1rj5v]

15:03 this is not at all the case. Humans use two eyeballs on a slowly moving gimbal attached to a meat based neural net to drive. The infrastructure today is already built for NN + vision. Nothing more needs to be added for a synthetic NN + 360 vision to drive a car far, far safer than humans.

[@airheart1]

Just needs A LOT of data and learning. And as Sandy stated.. latency of processing what is seen, and then decision making times.. is the key. And beyond just decision making times, is the key part of Neural nets correctly understanding every situation and making correct decisions.. no small task

[@user-to2rf1rj5v]

@@airheart1 Computer based NN's and cameras have waaaay less latency than human NN's + eyeballs. It isn't an easy problem at all, which is why it's taking even Tesla (who has the only correct approach) many years to get there. But they will get there.

[@3nityC]

When practical industrial expert meet theoretical credential expert

[@surferdude4487]

I suppose that at some point in time, vehicles will be equipped with transponders that chirp out: velocity, position, control inputs and vehicle footprint. That would save a lot of computing power when visual based systems won't have to infer that information.

[@James_Ryan]

15:02 Couldn't agree more with this, and let's not forget that human drivers (especially visitors from out of town) will also benefit from city layouts that don't have hidden dangers or unexpected layouts...

[@surferdude4487]

Yeah, sure, good luck with that.

[@theotherguy11701]

Thanks Sandy! Great example comparing the struggles with integrating traffic lights and the necessity of public supporting structures for self driving tech. Maybe it won’t be necessary but couldn’t hurt to try. Seems like there’s some confusion in the comments: the purpose of this company is to develop a testing system that evaluates existing self driving tech. Was wondering why most of this can’t be done entirely in software simulation? Couldn’t the hardware etc be simulated and then do thousands of scenario iterations?

[@dogtato]

i'd say there's value in doing it irl, but then they have simulated cars which i guess skips over the labeling step

[@jamesbell6182]

With all of the challenges of self drivers, maybe all of the autos in a confined area should be able to communicate with each other. As to what hazard are in the area , ie first responders, police activity or road construction

[@necbranduc]

10:40 - although this seems impossible to solve in the beginning and a very difficult way to find multiple examples of rare situations, Tesla has a solution for this. The simulated 3D environment where they can generate any type of vehicle driving on the road along with very edge-case scenarios, like a family with a dog running on the freeway, is what is going to solve most of these situations that are very rare in real life. However, with a computer that generates real-life-fidelity scenarios, you can balance out the dataset from which these models learn. PS: please don't tell anyone this secret

[@nroose]

Two problems with simulations: the end result of a machine learning process has less fidelity than the simulation in the first place; the rare situations are rare but there are many of them and we can't really think of all of them, or even most of them.

[@cherubin7th]

They showed this at AI day and it a common practice. Their biggest problem (and of every one else too) is that neural networks don't generalize well enough. It needs more zero shot ability.

[@CHIEF_420]

🎓

[@necbranduc]

@@cherubin7th Exactly.

[@vermontsownboy6957]

"Curse of rarity"? IN Tesla-world, it's called "edge cases". Tesla feeds these scenarios through their AI super computer to train their neural net.....data from billions of miles of Hardware 3-derived situations. U of M has well outlined the self-driving challenges, but Tesla set to work resolving this years ago!

[@bettyswallocks6411]

I used to live in Oxford, England. I can confirm that the department responsible for the local roads have never been known to pass up an opportunity to install a new set of traffic lights, regardless of need.

[@MsAjax409]

I felt like I was in AV kindergarden watching this video compared to graduate school watching Tesla's Autonomy Day presentation.

[@terrysullivan1992]

Somewhat ho hum stuff. These folks are way behind Tesla and on the wrong track. They need to test the latest Tesla Beta.

[@budgetaudiophilelife-long5461]

THANKS SANDY AND ALL …🧐 IT’S VERY INTERESTING AND WORTHWHILE….IMHHO…we look 👀 forward to seeing the results 🤗😍😍😍

[@rule1capital]

Without even trying or intending to, this video showed how embarrassing this entire effort is… the virtual crash and virtual traffic in the real car was the dumbest thing I’ve seen…

[@HansWeberHimself]

It might be the dumbest thing you have ever seen, but it is actually a very sophisticated simulation that has a lot of highly valuable applications. I don’t know where you’re coming from, but for me as a scientist (biology) this is extremely sophisticated research and will certainly be elemental to government agencies to evaluate novel approaches to self driving cars. It’s easy to be a fan boy, my friend. Anything else is rather difficult.

[@rule1capital]

@@HansWeberHimself This is pre-school level if you have ever driven in a Tesla with free Autopilot, let alone FSD Beta... if you are simulating other traffic, on the easiest possible fixed course, why even drive in a real car, when you can just simulate everything? Should I go "kid gloves" on them and say good job, here's your participation trophy... Being on Munro Live is like entering the Coliseum, be prepared to be judged against the best... It's nothing personal, but paint me a plausible scenario where this hardware/software catches Tesla and wins autonomy?

[@samyg123]

@@rule1capital considering MCity is trying to validate autonomous vehicles, this is actually a very elegant solution. Keeping surrounding vehicles/objects virtual allows for everything to be reused after a “crash”. Furthermore it is easy for MCity to change the conditions so OEMs don’t just design a system to pass a specific course. Watching them move on an empty track looks pretty goofy though. In reality they could probably do most of this either entirely virtually or with a multi axis dynamometer.

[@7sq]

Yes

[@LawpickingLocksmith]

Awesome! I tend to side with George Hotz when it comes to interactions between humans and AI. Endless work needed!

[@sa5cha63]

My sunup would be , thank you all for the learning experience and for giving me the awareness that humans and their machines should also be fully included into the Tesla approach experience to use a 🤖 🚘 in the near future. Special thanks 🙏 to Sandy for making the trip and the hosts to show us what they already invented. Take into consideration that soon there could be a fully modular wheel outdoor chair 🪑 which has all the systemic and modular needs inbuilt available for all car manufacturers, because not everyone will have a special Tesla (which could transport not just humans , but their personal mobility units , too) 🙏☮️🕉

[@Emm2004]

What is this virtual testing method any good for? This system is underwhelming besides the plethora of high end sensors

[@babaluto]

For not being immersed in to the FSD technology, this video did an excellent job of pointing out a factual perspective on where FSD is and what it will most likely take to get to level 5. Seems to me that Tesla may take a not so short term loss on how much they've had riding on the product being brought to market, sooner than realistically possible.

[@Clearphish]

I see that the flute from Rich Rebuilds has made an appearance! 😃

[@user-to2rf1rj5v]

24:08 They should at least be using a superior open source system, from Comma AI

[@ConservatEV]

I am a police records manager. We’ve had more crashes in our jurisdiction than there are days in the year… over 90% we’re off-roadway crashes. Parking lots, docks, etc. I didn’t see a large parking lot in the testing facility. It may seem basic, but that is where most minor incidents happen by far.

[@user-to2rf1rj5v]

18:00 this is absurd. I get why people jump to this conclusion. But think it through: humans are not connected via 4G LTE in their brains to sensors embedded in the infrastructure (thankfully, because we all know how bad it is when a phone call drops or sketches out). Solving vision with NN + 360 cameras is the ONLY way to drive orders of magnitude better than humans.

[@airheart1]

I don’t disagree. But it could accelerate the capabilities of systems to have more information from better viewpoints. I think it would be quite hard to make that information highly compatible with the multitudes of different systems in development. That is why city councils won’t waste their money. What information needs to be provided? In what standard format? Nothing exists that is standardized.. so whatever they do is likely to be a waste of money. The govt. has failed to take any leadership in this whole arena.. and they are the only ones who could address this. Tesla would never get approval to add proprietary sensors on public roads.. so they are forced to just work with their vehicles. If the govt. had an intelligent and useful DOT.. they could put out some standard sensor suites and transmit that info locally, nationwide.. and allow individual developers to tap into and use that info to their best ability. Could help at intersections for those dangerous edge cases. For instance, if emergency vehicles were mandated to report approaching intersections some 10 seconds out and that information was relayed THROUGH the correct intersection to all local vehicles they would be alerted prior to the intersection encroachment and be able to plan ahead better.. shit like that could be useful. As opposed to the AI is in the middle of accelerating into a ‘blind’ intersection when it finally sees what it identifies as an emergency vehicle coming.. then it has to decide.. what’s my correct action now? Continue or stop? Continue or STOP?? Omg?? And the people in the car are feeling that herky jerky shit of an AI in confusion.. solved by more communication of vehicles through better infrastructure. yeah I know the AI isn’t going to say OMG lol.. that’s the people in the car being bounced around

[@user-to2rf1rj5v]

@@airheart1 I don't disagree that some of this could be useful. Your emergency vehicle info is one of those situations. BUT, it cannot be the primary layer for reasons you laid out and more. Rather it would be useful as a secondary info layer. Primary must be NN's + vision, which is how humans drive today and how the infrastructure is already built out.

[@sudeeptaghosh]

Yup only deep learning AI can recognize such a complex environment and no we need to stop wasting time in connected systems like 5G or something

[@airheart1]

@@user-to2rf1rj5v absolutely agree.. it would just be supplemental info these AIs could learn to use to those few seconds ahead planning. But of course, what the cameras see in real time is the priority for safety

[@helmutkrahn9337]

@@airheart1 "..hard to make information compatible with different systems.." Road signage and traffic lights are pretty standardized, nationally at least... Which, admittedly, has taken many decades. Agreed, government should step up here; altho I have my doubts in view the current EV charger debacle.

[@pazznecht]

Would it make sense to add a shoulder/head rest for the wheel chair occupnat to protect them from more damage?

[@rickkay9548]

interesting that the white car system has exact knowledge of its own sim environment and all the pre-scanning and it still is a cluster f. Would love to see it in a city to city scenario to check actual perception

[@batchint]

something I came across some months ago was the diamond junction.. and some of which 'not just bikes' is super intelligence

[@shazam6274]

Another fine example of Government and Academia waddling around in a pool of money and bureaucracy. Nothing shown here that will ever make any difference in real life, commercial, applications for autonomous or automated driving. It's all about getting and spending Grant Money, "publishing papers", and shuffling to a nice retirement. The one key item mentioned is that Street Infrastructure, sensing and communication with vehicles is essential to have a possible working system and adaptation. Sandy correctly pointed out that there is zero effort, not even even discussion, in local municipalities, county or state governments considering this essential requirement for automated driving. Not much optimisms for this happening for two reasons: 1) "Lack of money"*. They can't even fix the rotting roads with the hundreds of billions of $$ of taxes supposedly designated* for that purpose. Drive around the cities of Michigan for an excellent example of the absolute crap road maintenance (same in most large cities). If this elaborate $$ test roadway is designed to test automated driving systems on real world roads, why are there no giant potholes and cracked asphalt and concrete, along with misaligned curbs and other crumbling "features" of roads, but which are a feature of the roads nearby? 2) Government is lazy. One of the examples in this video was concerned with dealing with emergency vehicles. Over 50 years ago it was proposed that along with flashing lights and sirens, emergency vehicles should transmit a signal so that cars in the path were alerted to their oncoming paths. In the 80s I discussed this with a Motorola engineer. He said that it would be simple and cheap. Motorola had even developed prototypes and submitted a proposal to the FCC. The Government, i.e. the FCC, regulates the airwaves and has numerous protected frequency bands for a wide assortment of specialized uses. For context, at the time, Motorola was the source for >85% of all the radios and communications routing for police, fire and ambulances in the US and could have easily incorporated this in to the existing radios. Other Government agencies regulate basic requirements for motor vehicles (i.e. 2 red lights at the rear of a very specific color of red). Just like seat belts, they could have easily required cars to have a detector for emergency vehicles using this radio technology. This would have saved a lot of crashes, injuries and deaths over the decades. This has not only not been implemented but is still not even under consideration.

[@aurelio-reymilaorcabal9669]

One of the announcements made when GM closed the Oshawa plant was to turn it into a R&D Center for AV, not sure what came of it, and I live in Oshawa , but have not seen or heard of àny developments.

[@surferdude4487]

The Oshawa plant is opening up again to produce BEVs. I haven't heard specifics on what models.

[@terrysullivan1992]

Sandy, please post the slide of the confusing traffic signals you mentioned.

[@lukerobson3043]

Going on sandys point I think traffic lights will have some sort of WiFi or Bluetooth to tell automatic car that the lights are going to change in x seconds so they can speed up or slow down

[@FutureSystem738]

Yeah, this could “be a thing” in maybe twenty or thirty years, assuming a standard is agreed upon. (Therein lies the problem.)

[@marknevar8234]

If only someone had lots of data on driving situations.

[@FutureSystem738]

LOL :-)

[@x0Kamix0x]

We are many many many years away from building up our infrastructure to include technogy that communicates with automated vehicles. Unless you're China, there's never going to be an all out push/consensus across North America.

[@eddiegardner8232]

I think Sandy just had a virtual whiplash neck injury from a virtual rear end collision.

[@Martin-se3ij]

There will be a virtual lawyer along shortly.

[@mikezurawski]

All vehicles need to talk to each other for L5. This will negate all required fixed based safety systems. Aircraft all have TCAS. TCAS in all vehicles will fix most crashes.

[@mrbuckmeister]

I agree that the current state is closer to automated driving. It's like driving macros. To change lanes I don't need to check my blind spot, look for an open spot adjust speed and turn my wheel. I just click my turn signal and the computer does the other steps. Sandy nailed it. New things are 10% tech and 90% psychology.

[@Futurama10000]

And how much does this build cost. I'm sure a tad more than cameras

[@Aleric735]

Then maintain, repair, defend (cyber), and upgrade. This doesn’t scale. Dead.

[@cassidybb10]

Nice quip Sandy lol "eh virtual neck brace so...." lol

[@markwebb7179]

The incentive for a city to deploy automation assistance is there, if you get the technology right. Every city wants to encourage young workers to move to their city (to look 'cool'). Every city wants to encourage people to spend time downtown. If you have a technology that can link the city and suburbs together in a cool new way, you can sell that to some cities in a cost-benefit way. Not all cities will take you up on this, but you don't need all of them to. You need 2-4 'demonstration' cities - preferably mid-sized cities like Cincinnati or Des Moines - to generate buzz and demonstrate workability. Those cities get a boost from having the technology, so they spend a few million to install it. What about travel between cities? Seems like freeway driving is a much simpler problem, one largely solved already by Tesla.

[@Factoryseconds123]

Tesla still hasn't solved phantom braking

[@RayNLA]

Sandy goes back in time to the year 2011

[@0cer0]

This is so sad. Not only do they know and understand little of what Tesla is doing (and what Tesla can do already today), they are wasting their resources with a conventional "a lot helps a lot" approach, and are also on a conceptual wrong track. They're in for a rude awakening.

[@davidharris7249]

The approach of smart roads is likely to fail because getting reliable transmission between all the parties with be difficult since they all need to agree on protocol, noise, and deal with latency. Each car still has to deal with all its inputs, and collate them to decide what to do. Deciding on priority of inputs can lead to indecision. Tesla's approach is to generalize the problem, and collect the vast amounts of data needed, via the >1M fleet, and process it on a fast supercomputer (Dojo) to teach the neural nets.

[@MozOnBikes]

As someone who’s lived in Oxford for ages, I’m curious to where the crazy traffic light that Sandy is scared of could be?!? Headington hamburger roundabout possibly!?!

[@protostark5860]

I know the university has the best of intentions for trying to develop automated self-driving systems but this is just very sad compared to what Tesla, Waymo, and others are already developing. I feel that university research funding would be better used for other endeavors that are not being fully developed already by the private sector currently.

[@sava411]

they literally say in the video they are not developing an AV system.

[@Dularr]

Where do you think the corporations will get future employees?

[@grantpaterson7132]

Dumb comment

[@protostark5860]

@@Dularr Teaching core fundamentals and academic theory is the correct place for students to gain that type of experience. However, most modern streamline private companies will fully train new employees from the ground up to work effectively in their corporate R&D environment. Notice how Sandy was interested mostly in the self-wheel chair locking system and nothing else. The basis of breaking into AV as a new research topic was done over a decade ago they should move on to something else like VTOL or Quantum Computing.

[@protostark5860]

@@sava411 The driving scenarios they are creating and analyzing are peanuts compared to the amount of miles logged by Tesla and Waymo. Also, Tesla does extreme case generation testing also with more data, better funding, and the new DOJO pod training exopods they are building. The university should be researching something no one has invested significant money into yet that will be a magnitude order better than what exists instead. University research dollars should always be for pushing the envelope of new areas of research to create a foundation for the private sector to take those new core research discoveries and they run with it to develop it with their own billions of dollars of capital, not research grants, endowments, and tuition. University is expensive enough as it is.

[@dougk5456]

WOW!!! Really informative and important to the future of FSD. This reminds me of when the FAA was formed. You must have a standard that everyone must meet and that platform is used to protect the company as well as the public. I hate to suggest that the federal government get involved because many of these people are technically illiterate, however, This would push the local governments and make them step up to the plate to standardize highway markings and systems to communicate with the vehicle.

[@MikeCasey311]

We don’t need more Federal anything.

[@gibby7616]

15:00 At the very least, government needs to require all emergency vehicles have #V2X to communicate to any autonomous vehicles nearby

[@willinwoods]

Preferably all vehicles, period. At least in the long run. Maybe a hub for general information, like traffic jams ahead, plus something like sharing gps data with other vehicles within half a mile or so. It could even be anonymous; you only need to know there IS a vehicle in your vicinity, seldom exactly WHAT KIND it is.

[@halfbrain86]

17:08 dude, have you ever seen a video of the latest Tesla FSD version in action?

[@willinwoods]

The accessibility of that entrance is a sad sight to see 😂 granted, it's probably not a custom vehicle built with driver ergonomics in mind (like the Rivian/Amazon delivery trucks), but still.

[@DaveTimperley]

A camera on every corner has privacy problems because they will be tracking everything with in view. When humans drive, they don't have 360 degree visual awareness. We discard a heap of info that flows into our brains, and sort out what we need. Once a car leaves our forward line of sight, it becomes imaginary and we expect that the driver of that car will obey the same rules we are following and it becomes a 'predicted object'. We don't have full situational awareness at a traffic light. We use audio cues for warning of an emergency vehicle, and everyone stops moving, until we have a visual marker for that emergency vehicle. Maybe AV's need microphones as part of their equipment?

[@rickkay9548]

This "we need smart roads for ai cars" is why no one has heard of this company or their products. A computer running a proper AI stack doesn't need ANY externally available infrastructure beyond traffic control - neither do humans

[@DominicZaidan]

Wow. Shows how far ahead Tesla is and will be if they solve the AI challenge. No need to do anything with roads, no big ugly equipment all over the cars, affordable.

[@whattheschmidt]

The affordable part won't matter much when they solve it. The vehicles would be worth over 200K each. This is why it's likely they won't sell many vehicles at some point and instead have a huge autonomous fleet. I don't think the hardware is there just yet (I think computer / camera upgrades are needed on the current vehicles) but we are likely only a couple of years away.

[@jamesengland7461]

The wheelchair locking device is a great idea

[@user-to2rf1rj5v]

There will be no level 4 or level 5 vehicles coming out of a closed course track. They *must* be tested and trained in the real world to get anywhere close to L3, let alone 4 and 5. The slide he spends a lot of time on (12:54) shows exactly why. This entire Mcity thing is a waste of resources and time.

[@floxer]

I agree. Will only be useful if you want to test your 'control' hardware, like the stuff to electronicly steer and brake and how its behaves under extrem circumstences. Like forcing it to drive trough mud or taking a corner at high speed.

[@sava411]

no AV system is "trained" at a closed course track. All AV developers use data from real world (or real world sims) to train NNs.

[@user-to2rf1rj5v]

@@sava411 Then what is the point of a closed track at all? If you train with real world data, you test in the real world. You'll never, ever get to 3, 4, 5 without doing that.

[@sava411]

@@user-to2rf1rj5v Testing in the real world is dangerous and puts others at risk. The point of a closed track is to validate your solution for a "baseline" performance to have a better idea of what errors to expect when testing IRL. If you design a plane with CFD/CAD, do you think the design should be tested in a wind tunnel before being tested in the sky?

[@user-to2rf1rj5v]

@@sava411 It's not any more dangerous than a licensed driver on the roads today. Every Tesla being tested in the real world has a licensed driver behind the wheel who can take over or intervene EASILY at any time.

[@MrGlotzTV]

FSD has been vaporware for many years and will still be for many years to come! These systems still struggle with mundane (for humans) situations. Neural nets have their limits (like flexibility, memory and prediction). Was looking forward to it finally making it to market, but I no longer hold my breath.

[@johnhjic2]

Hello Sandy, Not sure if this make me happy or it just a university way. After years of having people like Lancaster University in UK and a number of other transport company report to me in just developing routing and scheduling systems. Including purchasing navigational road data from all around the wold. which is a total nightmare if you want it to contain any restriction i.e. bridge heights or weight or width restriction. Allowing different types of vehicles down. As you see heir different road signage, let alone that each county signs it roads in a different way. not just languages but in different places. Just trying to do it for the US will be hard enough but the world. I know I do not have an EV but I do think the Tesla as the right approach in using camera and not relying on map and GPS. As a large number of map are still base on getting mercator projection right and at time the GPS system does go down and new roads are built. This is all a very big problem, and Yes it will take someone to develop test for these systems and I am sure these people have good intent but... Keep well, keep safe and please carry on you amazing work.

[@squashduos1258]

Ever driven by a traffic light which is out of commission….it happened to me today and this is not a one time occurrence….

[@gmv0553]

It will never happen without Palantir software!

[@beck99o]

This is interesting but I was not impressed with their Simulation/real driving test software. They want to create edge cases by having cars do aggressive manoeuvres, surely you want these to be where the ego car reacts in a safe manner. Each one they did just had a sim car run into Ego and crash!? I expect Tesla's SIM software is years ahead of whatever these guys are using and also being able to get data from real world like tesla is a much better way of doing it also. The more I see of other ways of doing things the more I just think Tesla is years ahead.

[@chickenwinck]

Exactly what i got from this. From what I’ve seen here im sure tesla is 5 years ahead. Leaning toward 10 years

[@vermontsownboy6957]

I'm confused. What's the point here? What exactly is U of M actually doing in Level 4/5 driving automation? I see vehicles covered in clunky devices, presumably LIDAR. I hear Prof Liu explaining the self-driving challenges, but it's worth noting that Tesla was chewing on the answers to these issues YEARS ago. The professor doesn't seem to have confidence in what the Tesla FSD team is doing. Given Sandy's standard comments about vehicles needing radar or LIDAR, and other external signalling information, I don't think Sandy understands what Tesla is doing either. Given that the vehicle is testing on an extremely tightly controlled and well mapped environment, this demonstration was, respectfully, completely underwhelming.

[@7sq]

They are years behind alpabet or tesla ...

[@murbella7]

Put Elon to work on the wheelchair lock. It could work like skis. A spring-loaded claw would engage by simply pushing the chair onto the rail. An electro release would be needed to exit.

[@user-to2rf1rj5v]

6:30 I really wish people would stop using this piece of art to show how complex human driving is. It's ART folks. en.wikipedia.org/wiki/Traffic_Light_Tree

[@floxer]

thx for that nugget of info

[@techfixr2012]

A simple Iff for Emergency vehicles that automated vehicles can use, that part is sloved.

[@gamesdean2621]

When the power goes out to smart intersections…the apocalypse begins

[@jamescodiroli722]

If a Tesla were to be tested, how would the test administrators "inject" all of the surrounding elements into the test? At the start and throughout the test, it appears to me, the only vehicle on the track is the one being tested. How will they make Tesla's 8 cameras and ultrasound sensors ”see" those adversarial elements ?

[@andreandre1051]

👍👍

[@ilin76bb]

FSD BETA from tesla is according to this professor, who wants to save his workplace, is a driver assistance tool. sure. LIDAR? Wow cool - no one mention what happens when it rains......... (its deaded)

[@romankrucker]

All this very complex technology needed for robo taxi, we have already very good efficient robo transport called, trains, busses, trams lol

[@tylerharris5606]

26:24 What's going on with Sandy in the rear view mirror here?

[@andi91324]

I'm wondering why no one is talking about the safety related features coming from technologies like vehicle to vehicle based on wireless communication systems, where u could easily sort out a number of low case critical events in milliseconds instead of just relying of a very expensive infrastructure builds. This kind of technologies are working and supported by a large number of governments, institutions, companies,.... why the hell are they not yet mandatory in all vehicle?

[@RB-pi3jl]

Sounds like Tesla wins! M-City team admits it's all about the data and Tesla has the data.

[@ThisRandomUsername]

I tried to comment yesterday but it didn't stick. I'd love if you went to see the guys at Comma AI. They seem to have a really good product already in the market, and are open about how it works. They'll have some good insight.

[@Antecxp]

I don’t see cities putting in any big chunk of change into automated/self driving vehicle technology when there is no real solution yet. We can’t agree on a charging plug standard. When costs come down significantly on a proven system then I can see a standard being adopted because there will be mass adoption. You don’t need crumple zones, a lot of safety equipment when you have a vehicle/s that does not get into accidents. We will be traveling in plastic or cart board boxes. You won’t need lights at intersections if the vehicles slow down or speed up in order to go through intersections while missing other vehicles by millimeters.

[@rowenagrinsam8261]

They still at kindergarten level compared to Tesla.

[@TFB-chris]

They are not developing AI driving systems, they test them and do tests on subsystems for automated driving systems. Totally different book.

[@sudeeptaghosh]

@@TFB-chris but the target is same right to drive autonomously on public road ? If yes then comparison is inevitable agreeable ?

[@TFB-chris]

@@sudeeptaghosh no, it is total different thing. They Test and validate systems and make assumptions how improve automated driving in a more general matter. They don't compete with Waymo, Tesla or whatever

[@sudeeptaghosh]

@@TFB-chris yeah you are right Tesla will drive on real roads with real world with pets people other vehicles but these guys will drive on simulators which is more important.. sorry my bad

[@TFB-chris]

@@sudeeptaghosh you miss the point. Completely.

[@krisvandermeulen253]

I wonder if any of these academics have had a ride in a FSD Beta yet...

[@Dontslaythybroski]

saaandyyyyy!!!!!

[@MikeCasey311]

I didn’t hear a word about Neural Networks. Just as Neural Networks are champion chess and go players, eventually they will be better drivers than the worlds best human drivers. 🇺🇸

[@Matthew.Sirrom]

Acceptance lag is an issue but i thinks its getting less and less as the years go by look at how people are embracing space and space flight just because of space X..

[@steamtorch]

Want to validate in a hurry? Drive across Paris during rush hour. Car comes out intact without killing anyone, it's GTG.

[@CHIEF_420]

🚲 ⚡

[@aastar699]

I like what they are doing. All seems a little dated to me.

[@TheRocky3211]

Looking forward to Tesla ai day reaction video

[@diskdrive123]

You have to have cars do 99% of everything with there own vision. You'll never have 100% coverage of every intersection or complex interchange.

[@eprohoda]

Wow~evry top -see u around!;))

[@henrikmunk450]

I thought self driving was just around the corner, but what do I know. This seem like Waymo turtle speed, not doing the job this century. I'm amazed of the lack of tech, just arguing about complexity. Seems like the winners of this battle will be decades ahead. Peace.

[@cassidybb10]

This is so interesting love it. The combination of all the industries and cities to intro connect...what happens when the power goes out? We have high aspersions...but rely on too much. Just drive your car. Wasting too much money on automated drive systems. Dumb Dumb. Dumb.

[@Sal3600]

Curse of rarity=edge cases

[@ckennylin717]

Why isn't audio input used in autonomous driving? There are a reasonably limited number of siren types on emergency vehicles such that AI can be easily trained on them. Unlike humans, if a vehicle "hears" a emergency vehicle siren, it will take less time to triangulate the source and respond appropriately. Humans still need time to figure out where the siren is coming from, confirm visually, then respond.

[@davidrowewtl6811]

Developing a testing system is so old fashioned. We have big data on accident rates for humans. Once any FSD beats some arbitrary level of driver (not having accidents) say at >19/20 drivers then it will be ready for V1.0 release. All the regulators need to do is set the acceptable human level to exceed. The insurance industry will do the rest. Cheap, open, fast.

[@ericcamarda3534]

I am not enthralled with the vehicle-to-infrastructure communications path. Seems like too many potential points of failure. Seems to me the vehicle has to fully autonomous, with no communications necessary to a central control center or city infrastructure, for driverless cars to proliferate.

[@krisvandermeulen253]

Their approach is different from Tesla as they directly aim for level 4 and 5 whereas Tesla goes for full level 2. BUT Tesla achieves level 4 immediately after their full level 2 gets enough training. And they will add all other 4&5 systems cannot collect enough real road driven data.

[@nroose]

The problem with smart roads is that if an intersection goes down, or even has a bug, you get a huge problem.

[@cherubin7th]

True, every few weeks I encounter a traffic light that goes down.

[@robertgow1769]

Sandy does not look imporessed whatsoever. Looks like he wants to take them for a ride in a Tesla with FSD Beta.

[@techunboxer]

Yeah lol

[@Emm2004]

Unfortunately, yes

[@darbyelliott2890]

Agreed

[@protostark5860]

Exactly! This is my take away as well, Sandy is not impressed with what was presented.

[@timepstein274]

Unfortunately you have all missed the point completely. They emphasised that this company doesn't do self-driving - they *test* self driving systems provided by others and were clear that they were not impressed either with the open source self driving system in the car provided. However, Sandy was impressed with their testing facility.

[@JosiahAugenstein]

No, government needs to stay out of it. I want my car to work independent of data being provided by the government. What happens in small towns and rural areas. What happens when the power goes out... The cars need to be self sufficient.