Connect with us

Tech

Google makes robots smarter by teaching them limits – TechCrunch

Published

on

 


If you’ve used smart voice assistants such as Alexa, Siri, and what Google calls smart assistants, you’ve noticed that the technology is getting smarter every day. Google is waiting for you, Siri can speak in a gender-neutral voice, and Alexa can read your bedtime stories in the voice of your late grandmother. As we covered at last month’s Robotics event, robotics is also evolving by leaps and bounds. The gap between his two voice commands and autonomous robotics is huge for a variety of reasons. Last week, I went to Google’s robotics lab in Mountain View to see how that might change in the near future.

Teaching a robot how to perform repetitive tasks in a controlled space that humans are not allowed to enter is a difficult but more or less solved problem. Rivian’s recent factory tour reminded us of that, but industrial robots are ubiquitous in manufacturing.

General-purpose robots that can solve different tasks based on voice commands in human-populated spaces are much more difficult. “But what about the Roomba,” you might say, but everyone’s favorite robot vacuum is generally programmed not to touch anything other than or on the floor.

Table tennis is a game in which robots can easily self-judge whether a task was successful or not and learn from their failures. Here, one of Google’s robotics researchers is taking a break and showing us what’s what in robots. Image credits: Haje Kamps (opens in a new window) / TechCrunch(opens in a new window)

“You may be wondering why we play ping pong. One of the big challenges in robotics today is the intersection of fast, accurate and adaptable. There is.It’s not a problem.It’s ok in an industrial environment.But being fast, adaptable and accurate is a huge challenge.Ping pong is a really great microcosm of this problem.Accuracy and speed. You can learn from people you play with, and it’s a skill that you acquire through practice,” says Vincent, a prominent scientist and head of robotics at Google Research. Mr Vanhoucke told me. “It’s not a skill that makes you a champion overnight after reading the rules.

Speed ​​and accuracy are one thing, but what Google is really trying to decipher in the Robot Lab is the intersection of human language and robotics. We have made some impressive leaps in the level of robots understanding natural language that humans might use. “If you have time, could you bring me a drink from the counter?” is a very simple request that you can ask a human being. But to the machine, this statement seems to combine a lot of knowledge and understanding into his one question. Let’s break it down: Sometimes the word “wait a minute” is just meant metaphorically and doesn’t make any sense at all, or it’s actually a request for the robot to finish what it’s doing. Sometimes it is. If the robot is too literal, the “correct” answer to “can you bring me a drink” might just be the robot saying “yes”. I can do it. Make sure you can grab a drink. But as a user, you didn’t explicitly ask the robot to do that. And if we’re being very pedantic, you didn’t explicitly tell the robot to bring you a drink.

These are some of the problems Google is tackling with its natural language processing system. The Pathways Language Model — or PaLM among friends: Instead of literally doing what humans say, it processes and absorbs exactly what humans actually want.

The next challenge is to recognize what the robot can actually do. The robot might understand perfectly if you ask it to grab a bottle of cleaner from the top of the refrigerator, which is kept safely out of reach of children. The problem is that robots can’t reach that high. The big breakthrough is what Google calls “affordances”: what robots with some success can actually do. This can range from simple tasks (“move forward a meter”), to slightly more advanced tasks (“go find a coke can in the kitchen”), to complex and multifaceted tasks that require the robot to demonstrate considerable comprehension. May include step actions. one’s abilities and the world around them. (“Hmm, I spilled a can of Coke on the floor. Can you mop it up and bring me a health drink?”)

Google’s approach uses the knowledge contained in the language model (“Say”) to determine and score actions that are useful for high-level instructions. It also uses affordance functions (“Can”) that enable real-world grounding and determine possible actions in a given environment. Using the PaLM language model, Google calls it PaLM-SayCan.

Google’s robotics lab uses many of these robots from Everyday Robots. These chaps are doing their well-deserved R&R (rest and recharge) and even learned how to plug in to recharge.Image credit: Haje Kamps(opens in a new window)/TechCrunch(opens in a new window)

To solve the more advanced commands above, the robot needs to break it down into several individual steps. An example is:

come on speaker Look on the floor, find the spill and remember where it is. Search drawers, cabinets, and kitchen counters for mops, sponges, or paper towels. If you find a cleaning tool (there is a sponge in the drawer), pick it up. Close the drawer. Go to Spill. Clean up any spills while monitoring that the sponge can absorb all the liquid. If not, wring it out in the sink and come back. Once the spill is clean, squeeze the sponge again. Turn on the faucet, rinse the sponge, turn off the faucet, and squeeze the sponge one last time. Open the drawer, put the sponge away, close the drawer. Identify the drinks you have in your kitchen and somehow decide which ones are “healthier” than Coke. Find a bottle of water in the refrigerator, pick it up, and bring it to the person who asked for it. The person may have moved after asking the question. He sinks 14 times. I thought it would be a great idea to use a small kitchen sponge to wipe off 11 ounces of liquid instead of using paper towels.

Anyway — I’m joking here, but you get the point. Even instructions that sound relatively simple can actually involve numerous steps, logic, and decisions along the way. Have you found the healthiest drink, or is your goal to have a healthier drink than Coca-Cola? Get your drink first, then figure out the rest of the task Does it make sense to allow humans to quench their thirst in between?

This screenshot shows how the robot evaluates the query and decides what to do at each step. In this case, the robot has to find the sponge, and we determined that there is a high success rate in finding, identifying, and manipulating the sponge. It also shows that this particular robot is very good at “going to the drawer”, which we conclude is not very useful in this example. Image credit: Google

The key here is to teach robots what they can and cannot do, and what makes sense in different situations. A tour of Google’s robotics lab shows more than 30 different robots, both Everyday Robots and dedicated machines, playing ping-pong, catching lacrosse balls, building blocks, opening refrigerator doors, and more. and learned to behave “politely” in the same room. universe as a human.

Nice catch! Image credit: Haje Kamps (opens in a new window) / TechCrunch

An interesting challenge facing robotics is that language models are not inherently based on the physical world. They’re trained on a huge text library, but the text library doesn’t interact with the environment and they don’t have to worry too much about things going wrong. It’s kind of funny that when you ask Google to show you the nearest coffee shop, Maps mistakenly shows him 45 days of hiking and his 3 days of lake swimming. In the real world, stupid mistakes have real consequences.

For example, if you see the prompt “I spilled my drink, can you help me?” the language model GPT-3 responds: It makes sense: for some clutter, vacuuming is a good choice, and it’s no surprise that the language model associates vacuuming with cleaning. If a robot actually did that, it could fail. Vacuum cleaners are not suitable for spilled drinks and do not mix water and electronics. So at best your vacuum will break, or at worst your appliance could catch fire.

Google’s PaLM-SayCan enabled robots are placed in a kitchen setting and trained to become better at different aspects of helping in the kitchen. When given instructions, the robot tries to make decisions. “How likely is it that I’ll be successful at what I’m trying to do?” and “How likely is this thing to help?” getting smarter

The triumphant return of the sponge-picking robot. Image credit: Haje Kamps (opens in a new window) / TechCrunch

Affordances, the ability to do something, are not an either/or. Balancing three golf balls on top of each other is very difficult, but not impossible. Opening a drawer is almost impossible for a robot that is not shown how the drawer works. work. As suggested by Google, an untrained robot may not be able to grab a bag of potato chips from a drawer. But a few instructions and a few days of practice can greatly increase your chances of success.

Of course, all this training data is scored as the robot tries. Sometimes robots “solve” tasks in surprising ways, but in reality it may be “easier” for them to do so.

By decoupling the language model from affordances, robots can “understand” commands in different languages. The team has also demonstrated it in the kitchen. Head of robotics, he is when Vincent Vanhoucke asks the robot for a can of Coke in French. “We learned language skills for free,” the team said, and the neural networks used to train robots were enough to open new doors (both literally and figuratively) in accessibility and universal access. I emphasized that I have a lot of flexibility.

Most robots that touch, open, move, or clean things don’t typically operate this close to humans. Although we were advised to keep our distance, the researchers seemed very familiar with robots operating autonomously within inches of an unarmed human body.Image credit: Haje Kamps (opens in a new window) / TechCrunch

Robots and technology are neither currently available nor intended for use in commercial products.

“It’s all research right now. As you can see from our current skill level, we’re not quite ready to deploy it in a commercial environment yet. We’re a research outfit and love working on things that don’t work,” Vanhoucke says. says. “That’s kind of the definition of research, and we keep pushing. We like to work on things that don’t have to scale. It’s a way of letting us know how we scale, so we can see trends for where things are going in the future.”

It will take some time for Google’s robotics lab to grasp what the long-term commercial implications of its experiments will be, but even the relatively simple demo shown in Mountain View last week , natural language processing and robotics both win as teams at Google build deeper skills, knowledge and vast datasets on how to train robots.

Sources

1/ https://Google.com/

2/ https://techcrunch.com/2022/08/16/google-robotics/

The mention sources can contact us to remove/changing this article

What Are The Main Benefits Of Comparing Car Insurance Quotes Online

LOS ANGELES, CA / ACCESSWIRE / June 24, 2020, / Compare-autoinsurance.Org has launched a new blog post that presents the main benefits of comparing multiple car insurance quotes. For more info and free online quotes, please visit https://compare-autoinsurance.Org/the-advantages-of-comparing-prices-with-car-insurance-quotes-online/ The modern society has numerous technological advantages. One important advantage is the speed at which information is sent and received. With the help of the internet, the shopping habits of many persons have drastically changed. The car insurance industry hasn't remained untouched by these changes. On the internet, drivers can compare insurance prices and find out which sellers have the best offers. View photos The advantages of comparing online car insurance quotes are the following: Online quotes can be obtained from anywhere and at any time. Unlike physical insurance agencies, websites don't have a specific schedule and they are available at any time. Drivers that have busy working schedules, can compare quotes from anywhere and at any time, even at midnight. Multiple choices. Almost all insurance providers, no matter if they are well-known brands or just local insurers, have an online presence. Online quotes will allow policyholders the chance to discover multiple insurance companies and check their prices. Drivers are no longer required to get quotes from just a few known insurance companies. Also, local and regional insurers can provide lower insurance rates for the same services. Accurate insurance estimates. Online quotes can only be accurate if the customers provide accurate and real info about their car models and driving history. Lying about past driving incidents can make the price estimates to be lower, but when dealing with an insurance company lying to them is useless. Usually, insurance companies will do research about a potential customer before granting him coverage. Online quotes can be sorted easily. Although drivers are recommended to not choose a policy just based on its price, drivers can easily sort quotes by insurance price. Using brokerage websites will allow drivers to get quotes from multiple insurers, thus making the comparison faster and easier. For additional info, money-saving tips, and free car insurance quotes, visit https://compare-autoinsurance.Org/ Compare-autoinsurance.Org is an online provider of life, home, health, and auto insurance quotes. This website is unique because it does not simply stick to one kind of insurance provider, but brings the clients the best deals from many different online insurance carriers. In this way, clients have access to offers from multiple carriers all in one place: this website. On this site, customers have access to quotes for insurance plans from various agencies, such as local or nationwide agencies, brand names insurance companies, etc. "Online quotes can easily help drivers obtain better car insurance deals. All they have to do is to complete an online form with accurate and real info, then compare prices", said Russell Rabichev, Marketing Director of Internet Marketing Company. CONTACT: Company Name: Internet Marketing CompanyPerson for contact Name: Gurgu CPhone Number: (818) 359-3898Email: [email protected]: https://compare-autoinsurance.Org/ SOURCE: Compare-autoinsurance.Org View source version on accesswire.Com:https://www.Accesswire.Com/595055/What-Are-The-Main-Benefits-Of-Comparing-Car-Insurance-Quotes-Online View photos

ExBUlletin

to request, modification Contact us at Here or [email protected]