Connect with us

Gadgets

This self-driving AI faced off against a champion racer (kind of) – TechCrunch

Published

on

Developments in the self-driving car world can sometimes be a bit dry: a million miles without an accident, a 10 percent increase in pedestrian detection range, and so on. But this research has both an interesting idea behind it and a surprisingly hands-on method of testing: pitting the vehicle against a real racing driver on a course.

To set expectations here, this isn’t some stunt, it’s actually warranted given the nature of the research, and it’s not like they were trading positions, jockeying for entry lines, and generally rubbing bumpers. They went separately, and the researcher, whom I contacted, politely declined to provide the actual lap times. This is science, people. Please!

The question which Nathan Spielberg and his colleagues at Stanford were interested in answering has to do with an autonomous vehicle operating under extreme conditions. The simple fact is that a huge proportion of the miles driven by these systems are at normal speeds, in good conditions. And most obstacle encounters are similarly ordinary.

If the worst should happen and a car needs to exceed these ordinary bounds of handling — specifically friction limits — can it be trusted to do so? And how would you build an AI agent that can do so?

The researchers’ paper, published today in the journal Science Robotics, begins with the assumption that a physics-based model just isn’t adequate for the job. These are computer models that simulate the car’s motion in terms of weight, speed, road surface, and other conditions. But they are necessarily simplified and their assumptions are of the type to produce increasingly inaccurate results as values exceed ordinary limits.

Imagine if such a simulator simplified each wheel to a point or line when during a slide it is highly important which side of the tire is experiencing the most friction. Such detailed simulations are beyond the ability of current hardware to do quickly or accurately enough. But the results of such simulations can be summarized into an input and output, and that data can be fed into a neural network — one that turns out to be remarkably good at taking turns.

The simulation provides the basics of how a car of this make and weight should move when it is going at speed X and needs to turn at angle Y — obviously it’s more complicated than that, but you get the idea. It’s fairly basic. The model then consults its training, but is also informed by the real-world results, which may perhaps differ from theory.

So the car goes into a turn knowing that, theoretically, it should have to move the wheel this much to the left, then this much more at this point, and so on. But the sensors in the car report that despite this, the car is drifting a bit off the intended line — and this input is taken into account, causing the agent to turn the wheel a bit more, or less, or whatever the case may be.

And where does the racing driver come into it, you ask? Well, the researchers needed to compare the car’s performance with a human driver who knows from experience how to control a car at its friction limits, and that’s pretty much the definition of a racer. If your tires aren’t hot, you’re probably going too slow.

The team had the racer (a “champion amateur race car driver,” as they put it) drive around the Thunderhill Raceway Park in California, then sent Shelley — their modified, self-driving 2009 Audi TTS — around as well, ten times each. And it wasn’t a relaxing Sunday ramble. As the paper reads:

Both the automated vehicle and human participant attempted to complete the course in the minimum amount of time. This consisted of driving at accelerations nearing 0.95g while tracking a minimum time racing trajectory at the the physical limits of tire adhesion. At this combined level of longitudinal and lateral acceleration, the vehicle was able to approach speeds of 95 miles per hour (mph) on portions of the track.

Even under these extreme driving conditions, the controller was able to consistently track the racing line with the mean path tracking error below 40 cm everywhere on the track.

In other words, while pulling a G and hitting 95, the self-driving Audi was never more than a foot and a half off its ideal racing line. The human driver had much wider variation, but this is by no means considered an error — they were changing the line for their own reasons.

“We focused on a segment of the track with a variety of turns that provided the comparison we needed and allowed us to gather more data sets,” wrote Spielberg in an email to TechCrunch. “We have done full lap comparisons and the same trends hold. Shelley has an advantage of consistency while the human drivers have the advantage of changing their line as the car changes, something we are currently implementing.”

Shelley showed far lower variation in its times than the racer, but the racer also posted considerably lower times on several laps. The averages for the segments evaluated were about comparable, with a slight edge going to the human.

This is pretty impressive considering the simplicity of the self-driving model. It had very little real-world knowledge going into its systems, mostly the results of a simulation giving it an approximate idea of how it ought to be handling moment by moment. And its feedback was very limited — it didn’t have access to all the advanced telemetry that self-driving systems often use to flesh out the scene.

The conclusion is that this type of approach, with a relatively simple model controlling the car beyond ordinary handling conditions, is promising. It would need to be tweaked for each surface and setup — obviously a rear-wheel-drive car on a dirt road would be different than front-wheel on tarmac. How best to create and test such models is a matter for future investigation, though the team seemed confident it was a mere engineering challenge.

The experiment was undertaken in order to pursue the still-distant goal of self-driving cars being superior to humans on all driving tasks. The results from these early tests are promising, but there’s still a long way to go before an AV can take on a pro head-to-head. But I look forward to the occasion.

Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published.

Gadgets

Still can’t buy a Raspberry Pi board? Things aren’t getting better anytime soon

Published

on

Raspberry Pi Foundation

Shortages for lots of tech components, including things like DDR5 and GPUs, have eased quite a bit since the beginning of 2022, and prices have managed to go down as availability improves. But that reprieve hasn’t come for hobbyists hoping to get a Raspberry Pi, which remains as hard to buy today as it was a year ago.

The most recent update on the situation comes from Raspberry Pi founder Eben Upton via YouTuber Jeff Geerling—Upton told Geerling that Pi boards are subject to the same supply constraints since the last time he wrote a post about the situation in April. Around 400,000 Pi boards are still produced per month, and some of these are being earmarked to be sent out to consumer retail sites. But Upton says that most of these are still being reserved for and sold to commercial customers who rely on Pi boards to run their businesses.

In short, the update is that there is no update. Upton said in April (and nearly a year ago, when the company raised the price for a Pi board for the first time) that the Broadcom processors at the heart of older Pi boards have been particularly difficult to source, but that high demand had been just as big an issue. Demand for Pi boards increased during the pandemic, and there was no more manufacturing capacity available to meet this demand. Upton said a year ago that there were “early signs that the supply chain situation is starting to ease,” but backed-up demand could still explain the short supply even if the Pi’s components have gotten easier to buy.

If you’re trying to buy a Raspberry Pi in the US or other regions, the rpilocator spreadsheet can be a valuable resource, letting you know when various models are in stock for ordering at most common Pi retailers. According to the tracker, few Pi 4 boards of any stripe were available to buy through September, though, and if you’re looking for a specific RAM capacity, you will be stuck waiting even longer. Businesses that want to inquire about buying Pis are still encouraged to contact the business@raspberrypi.com email address to make their case.

Continue Reading

Gadgets

Google prototypes, open sources an extra-long keyboard with one row of keys

Published

on

Enlarge / Google Japan jokes that you can increase productivity by having two people type on the keyboard simultaneously.

Google Japan has a history of joke keyboard concepts that challenge common notions of computing input. The latest concept, the Gboard Stick Version, places every key in the same row, so hunting and pecking can take a more linear approach.

As shown in Google Japan’s YouTube video below, it appears Google Japan actually prototyped the lengthy keyboard. Google will not be mass-producing or selling it, but there are GitHub files available with open source firmware, circuit diagrams, and design drawings to build the keyboard yourself. The GitHub page is careful to note that “this is not an officially supported Google product.” Google Japan’s blog post from Saturday said you could make the Gboard Stick Version with a 3D printer.

Google Japan’s video for the Gboard Stick Version.

As designed, the keyboard is an extraordinary 5.25 feet (1,600 mm) longIf you think that’s lengthy, the company said the original prototype was 7.87 feet (2,400 mm) long. The keyboard uses 17 boards total, including 16 for mounting the keys and a control board.

Google Japan jestingly argues that this design is more convenient for cluttered desks, storage, and finding the right keys when typing. Google Japan’s video shows the keyboard with an alphabetical layout, as a user initiates touch typing by memorizing the distance of individual keys from the left border. Alternatively, it’s ‘easy’ to find P, for example, knowing that it’s the 17th key in from the left (the first key from the left is a search button, not A). Surely, this is all simpler than hunting and pecking up, down, left, and right on a traditional keyboard layout.

Google Japan’s page for the keyboard also suggests you can use it with a QWERTY or ASCII code layout.

Google Japan also pointed to the keyboard's single row simplifying cleaning.
Enlarge / Google Japan also pointed to the keyboard’s single row simplifying cleaning.

Many detailed use cases for this one-row keyboard are clearly jokes, from using it to measure your kid’s height and get items dropped behind the couch, to using it as a walking stick, or the “bug-fixing module,” aka net, that turns the keyboard into a bug catcher in case you encounter bugs when coding (get it?).

But one purported benefit we could actually get behind is how much personal space the keyboard naturally enforces in the office and beyond:

The keyboard looks to be a natural safe-distance buffer for those who have to return to the office.
Enlarge / The keyboard looks to be a natural safe-distance buffer for those who have to return to the office.

Google Japan’s outlandish keyboard concepts have been going on for years as a way to promote Google’s Gboard keyboard app. Past iterations have included the Gboard Teacup Version and Gboard Spoon Bending Version.

Continue Reading

Gadgets

The Pixel 6a for $350 ($100 off) makes for an incredible deal

Published

on

The Pixel 7 might be arriving this week, but if you’re not interested in any of that newfangled flagship stuff, have we got a deal for you! The Pixel 6a, Google’s cheaper, simpler smartphone, is on sale at Amazon and Best Buy for $100 off. That makes for a pretty incredible $349 price tag instead of the normal $449. If you don’t count bundling deals that require signing up for a new phone line, this is the lowest price we’ve seen the phone at.

The Pixel 6a is a dead simple 6.1-inch phone that covers all the basics. It has a 6.1-inch 1080p, 60 Hz display, 6GB of RAM, 128GB of storage, and a 4410 mAh battery. The phone has nearly every feature you could want, including an in-screen fingerprint reader, IP67 dust and water resistance, NFC, and Wi-Fi 6e compatibility. The biggest downside is that there’s no wireless charging. The headline feature is the flagship-class SoC, the same Google Tensor chip you get in the Pixel 6, but for a low (and now even lower) price. The Tensor won’t win any benchmark wars, but at this price, the only other comparable device is the iPhone SE.

As for why you might hold out a bit and get the Pixel 7 instead, you’d be getting a major screen upgrade if you buy the (probably $900) Pixel 7 Pro, which will pack a 6.7-inch 120 Hz display. You’d also be doubling the RAM (12GB) and upgrading the camera setup from the ancient IMX 363 sensor that powers the Pixel 6a. That would be more than double the price of this phone. though. Like we said in our review, if you’re not a phone snob (guilty), the Pixel 6a is the perfect phone for normal people.

Ars Technica may earn compensation for sales from links on this post through affiliate programs.

Listing image by Ron Amadeo

Continue Reading

Trending