Auto a guida autonoma: sviluppi, problemi e potenzialità

[pennellotref]

https://jalopnik.com/watch-a-waymo-av-get-freaked-out-by-traffic-cones-bloc-1846897182

 

Cita

Watch A Waymo AV Get Freaked Out By Traffic Cones, Block Traffic, And Evade Support Vehicles

 

Making cars drive themselves is very, very hard. Sure, we’ve come an awful long way since that Stanford-modified Volkswagen Touareg won the first DARPA Grand Challenge back in 2005, but we’ve still got a long way to go before full Level 5 autonomy. No matter what anyone says. In case you don’t believe me, why not check out this video of a Waymo robotaxi getting very confused by some traffic cones and causing all kinds of trouble, including escaping from Waymo’s own support team.

The video is from YouTuber JJRicks Studios, who has done many videos documenting his rides in Waymo’s fleet of self-driving, Level 4 (that means no driver needed, but limited to a set, known area of operation) converted Chrysler Pacifica minivans, which Waymo operates as a robotaxi fleet in Phoenix. Sometimes people throw eggs at them.

Waymo’s automated vehicles have had some incidents in the past, and used to employ safety drivers in the car, ready to take over if needed; Waymo has since eliminated the safety drivers in-car, and now the Pacificas are un-human’d (well, except for the passengers) with safety monitoring happening remotely.

In this video, JJRicks hails a robot ride, and, as seems to be his habit, documents the entire process on video. I suppose luckily for him (but what would be unlucky for anyone who wanted to, you know, get to their destination instead of getting YouTube clicks) things went pretty significantly off the rails.

It’s a long video, but he’s thoughtfully added a bunch of annotations in the timeline, so you can skip around to the good stuff:

 

It’s interesting to see how this all unfolds. For the first part of the video, the car behaves fairly well. It drives pretty reasonably, and executes some difficult maneuvers like making an unprotected left turn. It also takes some odd shortcuts through neighborhoods, perhaps because data suggested a less challenging path that way? It’s not really clear what the goals are there.

I should also mention that weather and visibility were about as good as one could hope for, and road conditions just about everywhere seemed excellent.

The thing that managed to completely stymie this driving robot were the same things that anyone who has tried autocrossing very likely has already abused: orange traffic cones.

Interestingly, the car’s vision system (which fuses camera, lidar, and other sensors) can clearly see the cones around the corner

So the car knows the cones are there, and seems to be attempting to plot a course that sort of avoids them. The presence of the cones has already confused the car, which remains paused at this point before the turn, blocking traffic. It’s not a big deal really, as it’s a side street and at least one other car is able to get around the baffled vehicle.

When the car is stuck and calls for help to Waymo’s roadside assistance, what I really like is how much that delights the passenger, who laughs and actually claps his hands in glee. Really, Waymo couldn’t have hoped for a better passenger to have in the car when it goes to hell like this.

Then, while waiting for the support vehicle, some algorithm in the car’s silicon brains must have arrived at a decision, because the Waymo car then made the turn, fear of cones temporarily overcome, but then stopped in the middle of the lane, another terrifying cone dead ahead.

Looking at the visualization again, we see the cones clearly marked, and the car mostly in the coned-off lane, but with its butt sticking out into the open lane of traffic.

At this point, a Waymo vehicle is on its way to get a human driver in the car to rescue it, as Waymo keeps support vehicles in the area of their cars to handle these types of situations. They claim they don’t assign support vehicles to AVs one-to-one, but JJRicks there seems skeptical of that.

Then, strangely, the AV reverses out of its stuck position (again, this is reversing into an active lane of traffic, not what I’d call a great driving idea) and attempts to move again, but then stopped, which only accomplished having the car block the entire lane, instead of just half the lane.

The support driver is almost at the car, but before they arrive, a road maintenance worker removes the cone from in front of the confused car, and comes to the window to ask what the hell is going on, basically, as other cars honk and line up behind the Waymo van.

Removing the cone must have given the car an idea, because before the Waymo support van could arrive, the car decides to make its escape, and starts driving again.

The Waymo van is covering some ground, and it seems like it “fixed itself” but then its old nemesis Orange Cone shows up, and the car stops yet again

........

 

[aboutdas]

Tesla Confesses to California DMV Self-Driving Tech is Overhyped (thetruthaboutcars.com)

 

Cita

Back in January, Tesla CEO Elon Musk said he remained confident that his company would be able to deliver a self-driving vehicle exceeding the capabilities of an average human pilot by the end of 2021. But this has become a tired excuse used almost reflexively by automakers for years, making the inevitable shifting of the goalpost so predictable that nobody even bothers to get upset anymore. Being lied to is just part of everyday living and the automotive sector is just one droplet in the overflowing bathtub of mendacity.

Unfortunately, organizations continue making the mistake of expecting to be given the benefit of the doubt as they continue repeating the same fables. We know they’re working on solid-state batteries and autonomous cars, but they’re hitched to these unrealistic expectations and completely fabricated timelines that draw our focus while they engage in slimier practices on the sly. While holding them accountable is often easier said than done, catching them in a lie is usually fairly simple. For example, the California Department of Motor Vehicles accidentally called out Tesla on the full self-driving (FSD) beta it’s been testing with employees. 

Introduced in the fall and touted by Elon Musk back in the summer of 2020 as a major breakthrough, FSD has been evolving for years. The latest version (Beta 8.2) just represents the most recent incarnation, which the California DMV learned has been massively overhyped. According to Reuters, the department stated that its March 9th conference call with Tesla indicated it was still at SAE Level 2 that constitutes partial automation but requires constant driver engagement — which isn’t all that novel and miles away from Level 5 (actual FSD).

“Elon’s tweet does not match engineering reality per CJ. Tesla is at Level 2 currently,” the DMV wrote in a memo about the meeting, which included real talk from Autopilot engineer CJ Moore.

From Reuters:

The memo was released by legal transparency group PlainSite, which obtained it under the Freedom of Information Act (FOIA).

“Tesla indicated that Elon is extrapolating on the rates of improvement when speaking about L5 capabilities. Tesla couldn’t say if the rate of improvement would make it to L5 by end of calendar year,” the memo said, referring to level 5 full autonomous technology.

The manufacturer has been subjected to enhanced criticism whenever one of its vehicles are crashed. Regulators and the public desperately want to know if the car was in Autopilot as if that makes some kind of difference when it’s still operating at Level 2. The real scrutiny should be in response to how often Tesla makes false promises or misleads its own investors/customers before the same standards are applied to other automakers engaging in practically identical behaviors. Let’s face it, it wasn’t all that long ago when just about every manufacturer on the planet stated that self-driving cars would be made commercially available before 2020.

Where are they?

It’s getting to the point where we should really start blaming ourselves for being such complacent losers for standing for repeat nonsense. The proof that we’ve been collectively had is there and there are plenty of videos online capturing just how ineffective some of the most advanced “self-driving” systems actually are. For example, YouTuber AI Addict tested Tesla’s FSD Beta 8.2 in Oakland, California, and the video (below) is a 13-minute sizzle real of close calls and the car making bewildering decisions.

Officially, Tesla has been careful to state that drivers are expected to keep their hands on the steering wheel and should be prepared to assume control of their Tesla at any time. Frankly, we wish the company (all automakers, really) would be a little more vocal about those limitations and put an end to these not-so-subtle hints (falsehoods) about the technical prowess of their products.
 

 

 

[jeby]

1 ora fa, aboutdas scrive:

Tesla Confesses to California DMV Self-Driving Tech is Overhyped (thetruthaboutcars.com)

 

 

No, ma dai, veramente???

[aboutdas]

No, ma dai, veramente???

Un modo gentile per dire "stiamo usando un nome truffaldino"

[Hybris]

Molti siti riportano che Tesla negli Usa inizierà a consegnare auto senza radar e farà quindi affidamento solo sulle telecamere.

Mi chiedo come sia possibile pensare di arrivare alla guida autonoma grazie al solo uso di telecamere.

Il sito di Tesla italiano riporta che il radar "fornisce ulteriori dati sull'ambiente circostante su una lunghezza d'onda ridondante in grado di vedere attraverso la pioggia forte, la nebbia, la polvere e persino al di là delle auto precedenti."

[aboutdas]

2 ore fa, Hybris scrive:

Molti siti riportano che Tesla negli Usa inizierà a consegnare auto senza radar e farà quindi affidamento solo sulle telecamere.

Mi chiedo come sia possibile pensare di arrivare alla guida autonoma grazie al solo uso di telecamere.

Il sito di Tesla italiano riporta che il radar "fornisce ulteriori dati sull'ambiente circostante su una lunghezza d'onda ridondante in grado di vedere attraverso la pioggia forte, la nebbia, la polvere e persino al di là delle auto precedenti."

Al tempo stesso, fuori dagli annunci del patron Musk, ci sono dei test per sensori LiDAR Luminar sulle Tesla.
https://www.theverge.com/2021/5/24/22451404/tesla-luminar-lidar-elon-musk-autonomous-vehicles

[Merak]

Senza parole

[v13]

metti che se ne accendesse uno, semaforo rosso a 120 in autostrada

[J-Gian]

Cita

[...] L'introduzione delle nuove regole inizierà da alcune aree circoscritte a partire dal settembre del prossimo anno, ma la novità più importante è che, in caso di incidente, la responsabilità non sarà del guidatore, ma dell'auto stessa (e, di riflesso, del costruttore).  [...]

 [...] il conducente dovrà prestare comunque attenzione alla strada, utilizzando i sistemi driverless "in maniera appropriata", solo su alcuni tratti stradali (aperti al traffico) e senza violare altre norme del codice. Sono inoltre definiti dei termini precisi per l'interazione umana con il veicolo e sono state stabilite delle manovre d'emergenza che le vetture dovranno essere in grado di completare autonomamente. Alcuni Adas, come per esempio il cruise control adattivo, non sono menzionati tra quelli utilizzabili, mentre altri, come quelli che gestiscono sia i movimenti longitudinali sia quelli laterali, rientrano nella nuova legislazione.  [...]

 

Articolo completo su: https://www.quattroruote.it/news/sicurezza/2021/07/05/guida_autonoma_francia_codice_della_strada.html

[4200blu]

Mobileye and SIXT Plan New Robotaxi Service

Autonomous ride-hailing service is expected to begin driverless pilot in Munich in 2022.

NEWS HIGHLIGHTS

• At IAA Mobility, Intel CEO Pat Gelsinger and Sixt SE Co-CEO Alexander Sixt announced a collaboration to begin offering a driverless robotaxi service in Munich starting next year
• Riders can hail one of the robotaxis using either the Moovit app or the integrated SIXT app for ride hailing, vehicle rental, car sharing and car subscriptions

 

 

A photo shows the Mobileye autononomous vehicle(AV) operating under the MoovitAV mobility services brand. It was revealed at IAA Mobility 2021 as the production vehicle and robotaxi ride-hailing service Mobileye will bring to market beginning with Germany in 2022 through a collaboration with Munich, Germany-based SIXT Group. The service will be operated by SIXT and leverage the demand-generation of Intel subsidiary Moovit, carrying ride-hail passengers in Mobileye-owned AVs equipped with Mobileye Drive, Mobileye’s full self-driving system. (Credit: Mobileye, an Intel Company)

MUNICH, Germany--(BUSINESS WIRE)--During a keynote at IAA Mobility today, Intel CEO Pat Gelsinger and Sixt SE Co-CEO Alexander Sixt announced a collaboration to begin offering autonomous ride-hailing services in Munich in 2022. The collaboration between Intel subsidiary Mobileye and SIXT, a leading international provider of mobility services headquartered in Germany, also aspires to scale driverless ride-sharing services across Germany and other European countries later this decade.

Riders will be able to access the service via the Moovit app as well as the SIXT app. The autonomous robotaxi offering will be included in SIXT’s holistic mobility platform ONE, which combines in just one app products for ride hailing as well as car rental, car sharing and car subscriptions. By integrating the services of cooperation partners like Mobileye, the ONE mobility platform gives SIXT customers worldwide access to more than 200,000 vehicles, 1,500 cooperation partners, around 1.5 million drivers and soon even robotaxi services.

More: Intel and Mobileye at IAA Mobility (Press Kit) | Intel CEO Predicts Chips Will Be More than 20% of Premium Vehicle BOM by 2030 (News Release) | Intel CEO Keynotes at IAA Mobility (Video Replay) | Mobileye Moves from the Garage to the Streets (News Byte)

The autonomous robotaxi option will be part of the ride-hailing service SIXT ride and was demonstrated during Alexander Sixt’s keynote walk-on. Mobileye also unveiled the vehicles – branded with MoovitAV and SIXT – that will be produced in volume and used for the robotaxi service in Germany. It is the first time Mobileye has publicly displayed its fully integrated self-driving system, known as Mobileye Drive™, in a vehicle that will be used for commercial, driverless ride-hailing services.

A recently enacted autonomous vehicle (AV) law permits driverless vehicles on German roads, allowing Mobileye robotaxis to begin early-rider testing on Munich streets in 2022. The fleet will thereafter move from test to commercial operations upon regulatory approval. “Germany has shown global leadership toward a future of autonomous mobility by expediting crucial AV legislation,” Gelsinger said. “Our ability to begin robotaxi operations in Munich next year would not be possible without this new law.”

The collaboration with SIXT is the first known commercial robotaxi service between a tech supplier and a mobility services provider. “With strong logistics and operational partners like SIXT, Mobileye can bring the promise of full autonomy to life in cities around the world,” said Mobileye CEO Prof. Amnon Shashua. “We’re delighted that Germany is a first mover.”

Alexander Sixt, Co-CEO of Sixt SE, added: “This strategic collaboration is the next step in expanding our integrated mobility platform ONE and underlines our company’s evolution towards becoming the industry’s leading provider of innovative and digital premium mobility. We are delighted to leverage the remarkable technology leadership of Mobileye to bring driverless mobility to customers in Germany and beyond.”

Mobileye will own the robotaxi fleet used in the Munich service, while SIXT will draw upon its established expertise in providing, maintaining and operating the fleet. The vehicles will include the MoovitAV service and SIXT branding once the service launches in Munich, to make it easy for customers to distinguish between traditional ride-hailing and the autonomous fleet vehicles.