...really, it's kind of a lame video game but hey.
Fucking moronic. This is the worst type of uninformed, fear-mongering bullshit. 1. No, the sensor can't see around the bus. NEITHER CAN YOU, SHITWIT. 2. The AV didn't see the red light. DUH. YEAH, THAT HAPPENED ONCE IN 2009. 3. The bike that is entirely out of the way, going away from the intersection was mis-identified as a pedestrian. SO FUCKING WHAT? 4. Oh, the vehicle heard the sirens. Great. I guess I can turn the Motorhead up on my sound system, then. Jesus this is stupid. Nobody anywhere has championed the view that you will be able to drive in everyday traffic TODAY with a solo AV. Every single instance of an AV operating in the wild is like the one in Arizona that is firewalled to specific area, or the Tesla's auto-drive thing. All legitimate efforts to build AVs are pursuing driving modes that eliminate ALL of the strawman problems shown in this moronic little game. FACT: AVs will not be operating as individual actors, with equivalent skills to humans. AVs will share information wirelessly via a network. So that school bus blocking the sensor? Doesn't matter. Because there's an AV that just passed through and mapped that spot before the bus showed up, and two more sitting behind the bus, and all the vehicles are sharing data. FACT: They will be operating in protected areas - like HOV lanes - or geofenced areas where they are trained to operate. There is no legitimate company currently professing any solution to driving in everyday traffic with a fully autonomous vehicle. Even Seattle closed an entire street all the way through downtown JUST for buses. THIS is the environment where AVs will operate. Not on the road behind Bubba in his lifted F350 rollin' coal through downtown at rush hour. Dear lord. The fossil fuel industry and automakers are absolutely panicked, to be bankrolling trite bullshit like this.
1) The sensor doesn't make the association between the thing that went behind the bus and the thing that might pop out from behind the bus. Humans can. 2) Here's Uber blowing clear through one in 2016 with a pedestrian in the crosswalk: 3) Uber killed a bicyclist just two years ago. 4) Perhaps more importantly, humans see other drivers hearing sirens. Emergency vehicle management is very much flock behavior. Google is the only company whose AVs are networked. Tesla? no. Uber? No. Ford? No. And while Google is geofencing their vehicles, Uber was testing in the entire state of Arizona. Tesla, meanwhile, unleashes it as "beta" and lets you think that your Model S is going to see the truck in front of it so that you can Youtube yourself watching Finding Nemo while it slams you under an 18 wheeler at 70.
The problem is being addressed at multiple different levels, and people are working together on solving this. Two programs that I am personally aware of: 1. Smart Intersections. Sensors around intersections to track all events within and around the intersection (pedestrians, vehicles, lights, etc.), and communicate with vehicles that are entering the intersection. Several US cities already have prototypes running. 2. All vehicle manufacturers buy their subsystems and technology from other companies. I happen to work for one of them that is installed in the majority of vehicles on the road today. Even if a Tesla doesn't speak to a Ford and doesn't speak to a Hyundai, they all use the same subsystems from the same company, and THOSE talk to each other, and to their Mfg. The future of vehicle autonomy and intelligence is coming, and diligent and serious people are working on the problem quietly, from many different directions. The fact that one junior journalist at the WaPo didn't happen to have the connections to speak to the people who are actually moving the ball forward in the background, is not a valuable measure of how seriously the top 5 automakers in the world are taking this. It's easy to write a shock-piece full of flimsy strawman arguments. It's hard to do quality research and provide real insights and valuable thought on a complex issue.
1) So what happens when a vehicle expecting "smart intersections" goes to Pahrump, NV? Is it going to shut down and insist the driver take over? Or is it going to say that it's running at 90% autonomy and the lawyers will argue that it's Pahrump's fault for not spending a million dollars per intersection upgrading their infrastructure for 4,000 cars a day? The difference between total adoption and partial adoption is failure. 2) Sure, all manufacturers subcontract their shit (Google buys their subcontractors) but they don't all buy from the same contractor. Same problem - partial interoperability isn't interoperability. Neither Tesla nor Uber's approach to autonomy is serious or diligent. They're both overwhelmingly irresponsible. They occupy that 80% of the problem that's easy and snow over the 20% that's challenging and people have died. I don't care whose sensors you're using - if the customer doesn't want to pay for ONSTAR, that car isn't connected. And unless the sensor manufacturer decides to chip in for a 3G contract for its own purposes, nothing on it is connected to anything else. We've now evolved from writing endless iterations of the Trolley Problem to discussing the actual technology at hand. I would argue that's massive progress. You can wave your hands and pretend that these are all trivial problems but FAA-approved airliners flew into the ground twice because one department at Boeing thought another department at Boeing was being overly cautious when they asked for two pitot tubes and yet another department at Boeing knew that if they mentioned the MCAS system the FAA wouldn't approve the plane. Uber disabled Volvo's existing safety system because they weren't clever enough to integrate its sensor suite. The self-driving prophecy has already been pushed off in no small part because that last 20% is a stone-cold bitch and articles like this illustrate the problems for people who don't even know what a LIDAR is. And you can cast aspersions on the author but it doesn't make the content any less factual.
I never said the problems were trivial. I said the articles covering the technology lacked any useful depth, and went for easy blood-and-guts scare tactics, rather than any real journalism. Something like 80% of the vehicles on the road today are operating with my parent company's electronic subsystems in them. Whether the customer subscribes to OnStar or not, many of these talk to the manufacturer or to the cloud, and transmit data and metrics that the mfg uses in a variety of ways. Hell, our boxes are installed inside of some engines. The AV space is undergoing a lot of research, testing, and innovation, by a very large number of businesses and researchers, who are all working different aspects of the problem from different directions. It's not a matter of IF; it's a matter of WHEN.
Lather, rinse, repeat for AI and fusion. Automation and sensing have been at the 80% threshold for years now. It's the last 20%. Do the articles covering the problems lack useful depth? Sure. But that's just Gell-Mann amnesia.It's not a matter of IF; it's a matter of WHEN.
I know you are being slightly glib with the AI and Fusion correlation, but I'ma push back on it a little bit because I think it's important. Vehicles are ever-changing, and the environment in which they operate is becoming increasingly more restricted (is that an oxymoron?), so it is easier to make both incremental and sweeping changes on the roads, than it is in AI or Fusion research. GPS apps. EV charging stations. Automatic windshield wipers. Auto-braking. Lane departure alerts. Computer vision systems. All of these technologies have gone mainstream in 20 years, and several of them are less than 10 years old. These can be iterated. They improve with infrastructural improvements, like modern cellular networks and data sharing. In contrast, both AI and Fusion have hit a technology wall, and any significant increase will be a once-in-a-generation innovation; not an iterative increase in capabilities. Even solar power is iterating rapidly and increasing capacity, while the "big leaps" in AI and Fusion are still beyond our current capabilities, and the path forward is more research science, rather than an Agile/Scrum development process. This is why I am more optimistic about where these technologies are today, and where they will be in 5 years, as opposed to the parallels you proposed. No big deal... just... I think properly scoping the problem space is important here...
Wasn't being glib at all. Totally down to shut you down. I've given this ample thought over the years And the problems I anticipated being solved by now aren't solved by now. 1) Vehicles are not ever-changing. four tires, steers-at-front has been insurmountably dominant since the Model T. From a vehicle dynamics standpoint, that's the whole can o'nuts. Add safety features, drive more or different wheels, it's still four corners two of which turn. Even in vehicles with four-wheel steering, the back wheels effectively drop out at speed. 2) GPS, as a national benchmarking system, has been accurate to max 4m 95% CI since Reagan declassified it in response to KAL007 getting shot down. Sure - we can play differential GPS signals with assorted wifi and other signals but those are dependent on local GIS. Go drive Whatcom County some time. Your GPS and your phone disagree because your GPS says everything is 25 feet west of where you are because the Whatcom County GIS coordinates disagree with the federal ones. Your phone goes "herp derp that can't be right" but it's also the one that doesn't know you didn't just take the off-ramp. 3) EV charging stations have nothing to do with driving, automatic windshield wipers are a capacitance relay, auto-braking and lane-departure alerts are a supplement to an alert and active driver and computer vision systems use the same fuzzy logic as every catastrophe on this page. The thing we miss about machine vision is it's heavily leveraged in environments where it can assist humans in making decisions and in every instance where it makes decisions on its own it freaks out about kangaroos. 4) Mainstream or no, the basic problem is the system of driving. You can automate a vehicle's speed, lane position, braking protocols and adverse weather decision tree but you can't automate a kid chasing a ball. There are simply environments that are a pain in the ass to drive and there are environments that are easy. It's easy to automate easy environments - cruise control was invented in 1948. Fuckin' Buckminster Fuller invented a guided bomb by training a pigeon to peck at a destroyer on a TV screen. But the problem is more than the sum of its parts and the level of autonomy necessary turns out to be much higher than initial assessments suggested. Your argument fundamentally boils down to I have experience in vehicular automation and solar, therefore I understand the problems in vehicular automation and solar whereas with fusion and AI I do not so I accept that they are impossible. But what you don't have experience with is large scale systems dynamics because nobody does. There was every reason to believe that DIA's baggage system would work. It did not. Despite the fact that we've had 35 years of improvements in computing, automation, tracking and all the rest, nobody has attempted it since. Underestimation of complexity. Complex architecture. Changes in requirements. Underestimation of schedule and budget. Dismissal of advice from experts. Failure to build in backup or recovery process to handle situations in which part of the system failed. The tendency of the system to enjoy eating people’s baggage. http://calleam.com/WTPF/?page_id=2086 I would like autonomous vehicles to be a reality. I would love for the 95% of commuters who dislike driving to tune the fuck out and surf Facebook stoned on their way to their jobby-jobs. But I also know that there's a lot more to it than anybody is willing to admit. Google, for their part, decided that nobody would attempt it without maps that were accurate to the inch, so they set out to make maps that were accurate to the inch. But both Tesla and Uber said "fukkit, let's kill some people" instead. We live in a world where Audi and Toyota nearly die from tiny control issues and where 124 people are killed by faulty ignition switches. Into this miasma, throw total vehicular autonomy. It's gonna be a while.Contributing factors as reported in the press :