Google’s WaveNetEQ fills in speech gaps during Duo calls

Google today detailed an AI system — WaveNetEQ — it recently deployed to Duo, its cross-platform voice and video chat app, that can realistically synthesize short snippets of speech to replace garbled audio caused by an unstable internet connection. It’s fast enough to run on a smartphone while delivering state-of-the-art, natural-sounding audio quality, laying the groundwork for future chat apps optimized for bandwidth-constrained environments.

Here’s how it sounds compared with Duo’s old solution (the first is WaveNetEQ):

As Google explains, to ensure reliable real-time communication, it’s necessary to deal with packets (i.e., formatted units of data) that are missing when the receiver needs them. (The company says that 99% of Duo calls need to deal with network issues, and that 10% of calls lose more than 8% of the total audio duration due to network issues.) If new audio isn’t delivered continuously, audible glitches and gaps will occur, but repeating the same audio isn’t ideal because it produces artifacts and reduces overall call quality.

Google’s solution — WaveNetEQ — is what’s called a packet loss containment module, which is responsible for creating data to fill in the gaps created by packet losses, excessive jitter, and other mishaps.

Architecturally, WaveNetEQ is a modified version of DeepMind’s WaveRNN, a machine learning model for speech synthesis consisting of autoregressive and conditioning networks. The autoregressive network provides short- and mid-term speech structure by having each generated sample depend on the network’s previous outputs, while the conditioning network influences the autoregressive network to produce audio consistent with the more slowly-moving input features.

Above: A schematic of WaveNetEQ.

WaveNetEQ uses the autoregressive network to provide the audio continuation and the conditioning network to model long-term features, like voice characteristics. The spectrogram  — i.e., the visual representation of the spectrum of frequencies — of the past audio signal is used as input for the conditioning network, which extracts information about the prosody and textual content. This condensed information is fed to the autoregressive network, which combines it with the audio of the recent past to predict the next sample in the waveform domain.

To train WaveNetEQ model, Google fed the autoregressive network samples from a training data set as input for the next step, rather than the last sample the model produced. This was to ensure WaveNetEQ learned valuable speech information even at an early stage of training, when its predictions were still low-quality. The aforementioned corpus contained voice recordings from 100 speakers in 48 different languages, as well as a wide variety of background noises to ensure that the model could deal with noisy environments.

Once WaveNetEQ was fully trained and put to use in Duo audio and video calls, the training was only used to “warm up” the model for the first sample; in production, WaveNetEQ’s output is passed back as input for the next step.

WaveNetEQ is applied to the audio data in Duo’s jitter buffer so that once the real audio continues after packet loss, it seamlessly merges the synthetic and real audio stream. To find the best alignment between the two signals, the model generates slightly more output than is required and then cross-fades from one to the other, avoiding noticeable noise.

Google says that in practice, WaveNetEQ can plausibly finish syllables up to 120 milliseconds in length.

WaveNetEQ is already available in Duo on the Pixel 4 and Pixel 4 XL — they arrived as a part of the feature drop in December — and Google says it’s in the process of rolling out the system to additional devices. It’s unclear which devices, however — we’ve reached out to Google for clarification and we’ll update this post once we hear back.

Source: Read Full Article


Researchers release data set of CT scans from coronavirus patients

In an effort to spur the development of systems that can quickly spot signs of the novel coronavirus, a team of researchers at the University of San Diego this week released a data set — the COVID-CT-Dataset — containing 275 CT scans collected from 143 patients with confirmed cases of COVID-19, which they claim is the largest of its kind. To demonstrate its potential, they trained an AI model to achieve an accuracy of 85% — accuracy they say could be improved with further model optimization.

It’s important to note that the U.S. Centers for Disease Control and Prevention recommends against the use of CT scans or X-rays for COVID-19 diagnosis, as does the American College of Radiology (ACR) and radiological organizations in Canada, New Zealand, and Australia. That’s because even the best AI systems sometimes can’t tell the difference between COVID-19 and common lung infections like bacterial or viral pneumonia.

However, folks like Intel’s Xu Cheng assert the systems from companies like Alibaba, RadLogics, Lunit, DarwinAI, Infervision, and might play a role in triage by indicating that further testing is required. “Simply put, it’s challenging for health care professionals and government officials to allocate resources and stop the spread of the virus without knowing who is infected, where they are located, and how they are affected,” he said last week in reference to a system from Huiying Medical that detects coronavirus with a claimed 96% accuracy.

In this case, the researchers didn’t collect the scans themselves. As they point out in a paper, the images used in much of the existing research on COVID-19-diagnosing systems haven’t been shared due to privacy concerns. Instead, they scoured 760 studies about COVID-19 published from January 19 to March 25 and used a tool called PyMuPDF to extract low-level structure information. From the structure information, they located the embedded figures within the studies, and they identified the captions associated with the figures.

Next, the researchers manually selected clinical COVID-19 scans by reading the captions. For scans they weren’t able to judge from the caption, they looked for text referring to the figure to make a decision. For any figure containing multiple CT scans as sub-figures, they manually split them into individual images.

The team concedes that because the data set is small, training models on it could lead to overfitting, a phenomenon where the model performs well on the training data but generalizes badly to testing data. To mitigate the problem, they pretrained an AI system on the National Institute of Health’s ChestX-ray14 data set — a large collection of chest X-ray images — and fine-tuned it on the COVID-CT data set. Additionally, they augmented each image in COVID-CT by cropping, flipping, and transforming them to create synthesized pairs.

Trained on 146 non-COVID scans and 183 COVID-19 scans, the researchers report that their baseline AI system demonstrated high precision but low recall, which in this context refers to the ability of the model to find all the relevant cases within the data set. For the next step, the team says they’ll continue to improve the method to achieve better accuracy.

Concerningly, it’s unclear whether any of the researchers notified patients whose scans they scraped from the publicly available studies. We’ve reached out for clarification and will update this post when we hear back.

Source: Read Full Article


ModelOp helps enterprises deploy, monitor, and maintain AI models

ModelOp, a startup developing AI software and development services for enterprises, today announced that it has raised $6 million. It plans to use the capital to support demand for its products in a market that IDC anticipates will be worth $8 billion by 2022.

The term ModelOps refers to the process of cycling analytical models from data science teams to production teams in a cadence of deployment and updates, and it typically requires extensive domain knowledge on the part of the engineers involved. ModelOp’s platform aims to streamline this by cataloging models and automating deployment, monitoring, and governance processes across customers’ organizations.

Indeed, according to Algorithmia, nearly 55% of companies haven’t yet deployed a machine learning model, and a full one-fifth are still evaluating use cases or plan to move models into production within the year. That jibes with a recent study conducted by analysts at IDC, which found that of the organizations already using AI, only 25% have developed an enterprise-wide AI strategy. Firms responding to that survey blamed the cost of AI solutions and a lack of qualified workers, as well as biased data and unrealistic expectations.

With ModelOp Center Version 2, which launched in general availability today, developers can benchmark model performance on multiple platforms while ensuring compliance with regulatory requirements. The company’s team optionally works with customers to create blueprints and “industrialize” the use of AI across their organizations, assessing the current state of maturity and prioritizing recommendations based on critical needs. After establishing key metrics and processes and evaluating return on investment for the ModelOps investment, ModelOp develops a comprehensive roadmap for implementation.

To date, customers and ModelOp’s consultancy team have created models to predict delinquent payment behavior, municipal securities pricing information, and key economic events. ModelOp says its clients include five of the top 10 largest financial institutions as well as Fortune 500 manufacturers, insurers, and credit bureaus.

Valley Capital Partners led Chicago-based ModelOp’s latest fundraising round, which saw participation from Silicon Valley Data Capital. As a part of the deal, Valley Capital managing partner Steve O’Hara joined the company’s board of directors. ModelOp also expanded its executive team with three new appointments: Sheau-ming Ross, as chief financial officer; Mark LeMonnier, as vice president of software engineering; and Linda Maggi, as vice president of marketing.

“As enterprise Model Debt grows quickly, the emphasis is now on getting AI models out of pilot and into production, and this is driving rapid growth in the market for ModelOps,” CEO Pete Foley told VentureBeat via email. “With this latest round of funding, we’re well positioned to grow and ensure that our platform and delivery capabilities stay well ahead of competitive offerings and can accommodate the evolving regulatory landscape.”

Source: Read Full Article


Take your AI project from concept to concrete ROI

The AI guide that puts business analysts and data scientists on the same page

AI is proven to create new opportunities, increase operational efficiencies, and reduce costs for businesses. Why do so many AI projects fail before they deliver actual ROI?

Learn how to go from abstract business problem to concrete AI solution. Get business analysts and data scientists working together from start to finish. Get started with this free eBook to develop AI projects with real business impact.

Download now!

You’ll learn:

  • How to turn business problems into ROI
  • How supervised machine learning projects get done, from start to finish
  • Which type of Machine Learning path is right for you
  • Where projects go wrong — and how to course correct
  • And more!

    Source: Read Full Article


Best Xbox One Controllers (March 2020): Xbox Series X Compatible Controllers

Microsoft has consistently improved the Xbox One gaming experience since the console’s launch in 2013. Controllers have been a huge part of that; as we head into the launch of the Xbox Series X, there are tons of Xbox wireless controller options as well as wired ones, including the Xbox Adaptive Controller, multiple iterations of the Xbox Elite controller, third-party alternatives to the standard controllers, and more. Many of them also work as PC gaming controllers, and since the next-gen console will be backward-compatible, you don’t have to feel guilty about buying a great new controller to make your favorite first-person shooter, fighting game, or other favorite titles more comfortable.

When choosing the gaming controller that’s right for you, here are a few things you should consider.

Best Xbox One controllers with Xbox Series X compatibility

With the number of great Xbox controllers on the market, including high-end options, you can rest easy knowing that if you drop a large amount of money on a pad, you’ll be able to use it with Xbox Series X and the next generation of Xbox consoles, as well as a Windows device.

Of course, the basic Xbox One controller is perfectly suitable. It features excellent ergonomics, smooth triggers, and accurate analog sticks, and if you need to face someone 1-on-1 in a fighting game, its clicky D-pad will serve you well enough. However, some of the alternative options will give you customization features, a more accessible experience, or even an edge over the competition.

If you need to keep your thumbs on both sticks during a tense firefight, there are several controllers from Microsoft, Scuf, and Razer that feature remappable back paddles. If you want to relive the past with a hefty dose of nostalgia, the Xbox One has two excellent options in Hyperkin’s Duke and X91 pads. And if you’re looking to streamline the gameplay experience as, or for, someone with limited mobility, then Microsoft has an excellent, accessibility-focused controller.

Xbox Series X controller: What we know about the next-gen pad

We won’t have hands-on with the Xbox Series X controller until later this year, but we already know quite a bit about it. At the end of 2019, we spoke to Xbox head Phil Spencer for the Xbox Series X’s reveal. He said Microsoft learned a lot from the Xbox Elite controller and through fan feedback. Two significant design changes for the Xbox Series X controller came from that. The first is a new hybrid D-pad, while the second is a share button. Aside from this and some obvious cosmetic changes, not much else has been adjusted, as Spencer believes the Xbox One controller is already “pretty good.” And like every controller on this list, it’ll be compatible with the Xbox Series X and any Xbox One console.

Xbox Series X And Xbox One News

  • Xbox Series X Release Date Listed On Xbox Site, But Microsoft Says No Announcement Yet
  • Xbox Series X: Full Specs, Target Performance, And More Revealed
  • Xbox Series X: Release Date, Specs, Price, And Everything We Know
  • Why You Can't Pre-Order Xbox Series X Yet

We’ll keep this article updated as we test new controllers, leading up to the release of the Xbox Series X and throughout its lifespan. While you wait for the next-gen console, be sure to read through our list of the best Xbox One controllers that will work with the Series X. It’s important to note that the prices indicated below are each controller’s standard price and don’t reflect any discounts or fluctuations.

Quick look: The best Xbox One controllers in 2020

  • Xbox One wireless controller — starts at $50
  • Xbox Design Lab controller — starts at $70
  • Xbox Elite Series 2 controller — $180
  • Xbox Adaptive controller — $100
  • Scuf Prestige — starts at $160
  • Razer Wolverine Ultimate — $150
  • Hyperkin Duke — $41.16
  • Hyperkin X91 — $32.76
  • Razer Atrox fight stick — $200

For more gaming controller guides, check out our recommendations, check out our picks for the best PS4 controller, the best Switch controller, the best gaming keyboards, and best gaming mice.

And for more Xbox-related guides, check out our guide to buying an Xbox One, the best Xbox One games, our most anticipated Xbox One games of 2020, Xbox Game Pass,

Best value

Xbox One wireless controller | starts at $60/$70

The Good:

  • Stark improvements have made for an excellent standard controller
  • Affordable price
  • Several different editions to choose from

Of course, you can’t talk about the best Xbox One controllers without first talking about the standard on which everything else is based. The Xbox One controller has undergone a significant makeover since it was first released with the console back in 2013. For starters, the bumpers have been tuned to make them more ergonomic and easier to push, while the entire front faceplate of the controller is now one solid piece of plastic–the plastic around the Home button used to be separate from the rest of the pad. In addition to that, the controller now features a 3.5mm headphone jack and Bluetooth connectivity. Other small changes have also been adopted, making it increasingly hard to go back to any previous iteration of the Xbox One controller. Despite all the changes, it still requires two AA batteries.

Different editions of the controller boast unique properties, such as extra texture on the hand grips or triggers. If you want a unique design for your controller, Microsoft offers Xbox Design Lab, which gives you the ability to change the colour of almost every single part of the pad–you can also add an engraved message. No matter what direction you go with a standard Xbox One controller, you’re going to get a great pad that works well with every game on the console. | Mat Paget

Best overall Xbox One controller

Xbox One Elite Series 2 controller | $180

The Good:

  • Newly added grips keep controller firmly in your hands
  • Trigger stops automatically adjust sensitivity
  • Three profiles for controller customization
  • Adjustable analog-stick tension
  • New thumb-stick heads emulate Xbox 360 controller
  • 40 hours of battery life

The Bad:

  • Uncomfortable with all four paddles attached

The Xbox One Elite Series 2 controller is hard to beat. With its textured hand and trigger grips, refined trigger stops, and adjustable stick tension, it’s quite the step up from the original Elite controller. It features all of the same customization options as well, but instead of only two profiles, there are four–one of which returns the controller to its default settings. It also boasts a built-in battery that can last up to 40 hours and Bluetooth connectivity, which was introduced to the Xbox One’s controllers after the release of the Elite Series 1 controller. It also features charging via a USB-C cable rather than micro-USB, which is an improvement.

There is a downside, though. Microsoft’s layout for the Elite controllers’ back paddles is a bit uncomfortable, and while it’s not terribly difficult to get used to, I do find it hard to get my hands into a comfortable position with all four paddles attached. Thankfully, I don’t feel the need to have all four paddles attached–I’m perfectly content with two paddles for crouching and jumping in my shooters of choice. However, when companies like Scuf make controllers with a comfortable layout for all four paddles, it is slightly disappointing by comparison.

Despite this setback, the Elite Series 2 is a delight to use. The extra hand and trigger grip feels nice, and being able to adjust the stick tension is a huge positive. The tighter analog sticks feel great, and when paired with the larger thumb stick heads, it emulates the Xbox 360 controller except with the more comfortable ergonomics of the Xbox One pad. On top of that, there are now two trigger stop positions as opposed to one, and by default, the controller adjusts trigger sensitivity on its own–previously, you’d have to do this in the Xbox Accessories app on Xbox One or PC. The clicky home button also has a more premium feel when compared to the mushy-ness of the basic Xbox One controller. All of this makes the Elite Series 2 feel like a next-gen controller, perfect for using with the Xbox Series X. | Mat Paget

Best Xbox One controller for accessibility

Xbox adaptive controller | $100

The Good:

  • Works with a wide range of assistive devices
  • Extremely flexible customization
  • Officially compatible with Xbox One and PC
  • Works on Nintendo Switch with Bluetooth adapter

The Xbox Adaptive controller is much different than the rest of the control options in this round-up. It’s intended first and foremost as a device that helps those with limited mobility play games. It works with a wide range of assistive devices that users can plug in and assign to specific controller inputs to give them the ability to play any game on the two platforms.

It features 20 ports for you to plug joysticks, switches, buttons, and any other assistive device into–19 of those are 3.5 mm ports, while the other two are USB 2.0 ports. There’s also a 3.5mm audio output port for headphones or a headset. It’s compatible with both Xbox One and PC, and there’s quite the dedicated community behind it, discovering new ways to use the adaptive controller–you can even utilize a Bluetooth adapter to get it working with the Nintendo Switch.

The Adaptive controller may not be for everyone, but thanks to its ability to effectively bridge the gap between gamers with limited mobility and the games they want to play, there’s no doubt in my mind that it’s the most important controller on this list. And with a little savvy, you can make it compatible with Switch. | Mat Paget

Best back paddles

Scuf Prestige | starts at $160

The Good:

  • Extremely comfortable back paddles
  • 30-hour rechargeable battery life
  • Rubberized grip feels great
  • Interchangeable analog sticks

Scuf has been making Elite-style controllers for years (you can check out our review of the Scuf PS4 controller), and it shows. The company’s controllers are some of the most comfortable you can find, and the Scuf Prestige controller is a great example of this. The Scuf Prestige is very similar to a standard Xbox One pad, though it definitely feels different. The plastic is much smoother on the Scuf controller, and the backside features subtle, yet effective rubberized grip. It also features an interchangeable faceplate and a built-in rechargeable battery with 30 hours of life.

The Prestige excels most in its four back paddles, which are the most comfortable we’ve tested, thanks to their vertical alignment and distinct textures, shapes, and sizes. Remapping the extra buttons is also quite simple, though you’ll need to make sure to hang on to a small accessory to do so–the EMR (Electro-Magnetic Remapping) key. All you do is place the magnetic key on the back of the controller, then hold a specific paddle and specific button for at least one second. Once you’re done remapping your paddles, just remove the EMR key and you’re good to go.

The Prestige also comes with two interchangeable thumbsticks and adjustable built-in trigger stops. Like all Scuf controllers, you can customize and build your own exactly to your liking on Scuf’s website. It starts at $160, and you can choose the colour of each and every part, as well as remove the rumble motors if you want to go that route. Even for the price, you’ll be hard-pressed to find a better third-party Xbox wireless controller that’s also great as a wired controller. | Mat Paget

Best buttons

Razer Wolverine Ultimate | $160

The Good:

  • Excellent, clicky face buttons
  • Six programmable buttons
  • Interchangeable D-pad

The Bad:

  • Only usable with a wired connection

The Xbox One Elite and Elite 2 controllers are regarded as some of the best high-end, pro-style gamepads out there. But Razer has its own take on that design with the Wolverine Ultimate. It may not sound like a game-changer, but the best thing about the Wolverine is how great its buttons feel. The face buttons mimic the tactile nature of mouse clicks and feel more responsive as a result, and the analog sticks are buttery smooth and frictionless (which makes minuscule movements and precision easier).

Another key feature of the Wolverine is its six programmable buttons–four are on the back and two are near the shoulder buttons. Of course, this means you can map face button functions to these additional inputs and keep your thumbs on the sticks at all times. Those buttons are effortless to press down as well. You can also customize the D-pad to be used in a traditional four-way or rounded eight-way layout.

There are a few drawbacks with the Wolverine Ultimate, one being that it can only be used through a wired USB connection. It usually retails around $160 USD, which puts it up in the territory of an Elite controller, too. But if you’re looking for an Elite-style controller and can find it on sale, the Razer Wolverine Ultimate would be a great option. | Michael Higham

Best retro-style Xbox One controllers

Hyperkin Duke | $70

The Good:

  • Recreates classic Xbox feel
  • Includes modern touches like shoulder buttons
  • Feels great to use with big hands

The Bad:

  • Can be quite cumbersome for smaller hands
  • Only useable with a wired connection

The Hyperkin Duke was made purely out of nostalgia for the original launch Xbox controller from way back in 2001–it was this hulking gamepad with oddly shaped and offset face buttons. So there wasn’t much surprise when it was quickly surpassed by the Controller S, which became the standard design moving forward. But if you have larger hands, the Duke might be a better fit.

Hyperkin has made a name for itself by recreating retro gaming experiences through its wide range of hardware, and its Duke controller very much resembles the original Xbox controller in terms of size and button layout, but with a few modern touches. While original Xbox controllers had black/white buttons instead of left/right bumpers, Hyperkin incorporated small bumpers so the Duke would make sense for playing today’s games. The huge logo on the center of the controller is a screen that also acts as the home button–when you power on, the screen displays the old Xbox splash screen. Otherwise, it’s a faithful recreation of the Duke that now works through USB for Xbox One and PCs. | Michael Higham

Hyperkin X91 | $30

The Good:

  • Retro form factor
  • Features every button a regular Xbox One controller does
  • Great for retro-style or D-pad-focused games

The Bad:

  • Mushy triggers
  • Only usable with a wired connection

In addition to Hyperkin’s wide array of retro gaming hardware that lets you play old games easily, it has a few retro-inspired accessories. One of those is the Hyperkin X91, an Xbox One controller that’s scrunched down into a SNES-like form factor. Despite the small size, everything you need in a controller is there and it all works surprisingly well. From the analog sticks to the face buttons, the X91 recreates the full controller feel almost perfectly. One downside is that the triggers can feel a bit squishy as opposed to the smooth feel of the triggers on a DualShock or regular Xbox One controller.

And if you have a gaming laptop and play on the go often, the X91 is the perfect size for travel. Unfortunately, this controller only works through wired USB. While that means you don’t have to worry about battery life, having a thick cord connected can make it a bit clunky to have around. The form factor may also make it slightly more difficult to be precise with the analog sticks since there isn’t much you can grip to keep the controller steady. However, if you need a small, fully-featured gamepad for less intense games, the X91 is a fine choice. | Michael Higham

Best Xbox One fight stick

Razer Atrox fight stick | $200

The Good:

  • Easily moddable
  • Excellent 8-way stick
  • Great buttons
  • Sanwa parts
  • Removable USB cable

The Bad:

  • No right-stick control or L3/R3 buttons
  • No official PC support

If you’re looking for an Xbox fight stick that will last, then the Razer Atrox is the one you want. Not only will it be forward-compatible with the Xbox Series X, but it’s also fully moddable, which means you can replace the joystick and buttons as you wish–and it’s as easy as pushing a button to pop the Atrox open and access its various wires and components. Despite its modding potential, it’s more than ready to go right out of the box. The Sanwa joystick and buttons feel great and are satisfying to tap combos out on. The USB cable is also completely removable, making it easy to store inside the fight stick’s compartment.

The Atrox may be the best stick I’ve used for the Xbox One so far, but it’s not quite perfect. Unlike Razer’s Panthera Evo PS4 stick, the Atrox is not officially compatible with PC and does not feature a switch that lets you swap the joystick from D-pad to either analog stick or a way to press L3 or R3. The cases in which you need these inputs in a fighting game are rare, but needing a regular controller for character customization or anything else that uses these inputs is a little disappointing.

Thankfully, the Atrox makes up for this when you get into the action. I tested it with Dragon Ball FighterZ, Street Fighter 30th Anniversary Collection, Tekken 7, and Dead or Alive 6 and was very happy with how it performed across different styles of fighting games. If you’re looking for a great, future-proofed Xbox One fight stick, then the Atrox is an excellent one to go with. | Mat Paget

More Tech Picks From GameSpot

  • The Best Nintendo Switch Controllers You Can Buy
  • The Best VR Headsets In 2020
  • The Best Budget Monitors Under $200
  • The Best Gaming Laptops In 2020
  • The Best Gaming Keyboards In 2020

Source: Read Full Article


Google’s AI helps robots navigate around humans in offices

In a study published this week on the preprint server, Google and University of California, Berkely researchers propose a framework that combines learning-based perception with model-based controls to enable wheeled robots to autonomously navigate around obstacles. They say it generalizes well to avoiding unseen buildings and humans in both simulation and real-world environments and that it leads to better and more data-efficient behaviors than a purely learning-based approach.

As the researchers explain, autonomous robot navigation has the potential to enable many critical robot applications, from service robots that deliver food and medicine to logistical and search robots for rescue missions. In these applications, it’s imperative for robots to work safely among humans and to adjust their movements based on observed human behavior. For example, if a person is turning left, the robot should pass the human to the right to avoid cutting them off, and when a person is moving in the same direction as the robot, the robot should maintain a safe distance between itself and the person.

To this end, the researchers’ framework leverages a data set aptly dubbed Activate Navigation Dataset (HumANav), which consists of scans of 6,000 synthetic but realistic humans placed in office buildings. (Building mesh scans were sampled from the open source Stanford Large Scale 3D Indoor Spaces Dataset, but any textured building meshes are supported.) It allows users to manipulate the human agents within the building and provides photorealistic renderings via a standard camera, ensuring that important visual cues associated with human movement are present in images, such as the fact that when someone walks quickly their legs will be further apart than if they’re moving slowly.

Google Berkeley robot AI

For the above-mentioned synthetic humans, the team turned to the SURREAL Dataset, which renders images of people in a variety of poses, genders, body shapes, and lighting conditions. The images come from real human motion capture data and contain a variety of actions, like running, jumping, dancing, acrobatics, and walking, with adjustable variables — including position, orientation, and angular speed.

After the framework generates waypoints and their associated trajectories, it renders the images recorded by the robot’s camera at each state along the trajectory and saves the trajectory, along with the optimal waypoint. The trajectory and waypoint are used to train a machine learning model that facilitates reasoning about human motion.

In experiments, the researchers generated 180,000 samples and trained a model — LB-WayPtNav-DH — on 125,000 of them in simulation. When deployed on a Turtlebot 2 robot without fine-tuning or additional training in two never-before-seen buildings, the model succeeded in 10 trials by “exhibiting behavior [that] takes into account the dynamic nature of the human agent.” Concretely, in one instance, it avoided a collision with a human by moving in the opposite direction, and in another, it took a larger turn radius around a corner to leave space for a person.

Google Berkeley robot AI

The team says their framework results in smoother trajectories than prior work and doesn’t require explicit state estimation or trajectory prediction of humans, leading to more reliable performance. Moreover, they say the agent can learn to reason about the dynamic nature of humans, taking into account people’s anticipated motions while planning its own path.

“In future work, it would be interesting to learn richer navigation behaviors in more complex and crowded scenes,” wrote the coauthors. “Dealing with noise in robot state estimation will be another interesting future direction.”

Google isn’t the only tech giant pursuing autonomous robot navigation research. Facebook recently released a simulator — AI Habitat — that can train AI agents embodying things like a home robot to operate in environments meant to mimic real-world apartments and offices.  And in a paper published last December, Amazon researchers described a home robot that asks questions when it’s confused about where to go.

Source: Read Full Article


Miso Robotics deploys AI screening devices to detect signs of fever at restaurants

Miso Robotics, a startup developing robots that can perform basic cooking tasks in commercial kitchens, today announced that it has deployed new tools to its platform in CaliBurger restaurants intended to improve safety and health standards. The hope is to minimize the threat of infection for patrons and delivery workers during the COVID-19 pandemic, which has sickened hundreds of thousands of people worldwide.

In the coming weeks, in partnership with payment provider PopID, Miso will install a thermal-based screening device in a CaliBurger location in Pasadena, California, that attaches to doors to measure the body temperatures of people attempting to enter the restaurant. Before entering, the staff, delivery drivers, and guests will have to scan their faces, and if the device sensor detects the person has a fever, they won’t be allowed to enter the building.

Miso says that store owners will be able to opt into text messages alerting them that someone whose temperature reading is in line with health and safety standards is at the door, at which point employees will be able to open the door manually. Other settings will allow for security and privacy measures, including facial recognition linked to approved employees or residents.

“While the initial use case will be for CaliBurger, the technology could be applied to other buildings such as offices, homes and other public places of business to ensure the health and safety of those entering,” said a Miso spokesperson via email. They also noted that while all scan data is anonymized, it can be pooled together to analyze patterns in populations and regions and identify trends. “Wider adoption could serve to identify hot spots and pikes in the spread of COVID-19 and other transmissible viruses within populations and regions, serving as a source of information for officials seeking to traction initial sources of outbreak.”

Fever detection is on trend. Various companies have begun deploying cameras that detect people who may have COVID-19 by using thermal imaging, including Austin-based Athena Security. Ramco Innovation Lab in Singapore recently unveiled a facial recognition-based system with embedded temperature recording, and in China, authorities have reportedly used drones to detect signs of fever.

Miso also says it will also install physical PopID terminals so that guests can transact without touching a panel, using cash, or swiping a credit card — all of which can transfer pathogens. Customers will also be able to pay using PopPay on the CaliBurger website.

According to Miso, CaliBurger intends to implement these technologies in all of its restaurants.

Miso has long claimed that its Flippy robot — and Flippy’s successor, Miso Robot on a Rail (ROAR) — can boost productivity by working with humans as opposed to wholly replacing them. Miso AI, Miso’s eponymous cloud-based platform, orchestrates the prep of over a dozen food items, including chicken tenders, chicken wings, tater tots, french fries and waffle fries, cheese sticks, potato wedges, corn dogs, popcorn shrimp and chicken, and onion rings.

Miso claims that ROAR can prep hundreds of orders an hour thanks to a combination of cameras and safety scanners, obtaining frozen food and cooking it without assistance from a human team member. It alerts nearby workers when orders are ready to be served, and it takes on tasks like scraping grills, draining excess fry oil, and skimming oil between frying as it recognizes and monitors items like baskets and burger patties in real time. Plus, it integrates with point-of-sales systems (via Miso AI) to route orders automatically and optimize which tasks to perform.

Source: Read Full Article


Uber’s Enhanced POET creates and solves AI agent training challenges

In a paper published this week on the preprint server, Uber researchers and OpenAI research scientist Jeff Clune describe an algorithm — Enhanced Paired Open-Ended Trailblazer (POET) — that’s open-ended, meaning it can generate its own stream of novel learning opportunities. They say it produces AI agents capable of solving a range of environmental challenges, many of which can’t be solved through other means, taking a step toward AI systems that could bootstrap themselves to powerful cognitive machines. Picture enterprise AI that learns an objective without instruction beyond a vague task list, or cars that learn to drive themselves in conditions they haven’t before encountered.

It’s in some way an evolution of Uber’s work in games like Montezuma’s Revenge, which the company detailed in late November 2018. Its Go-Explore system, a family of so-called quality diversity models, achieved state-of-the-art scores through a self-learning approach that didn’t require human demonstrations.

As the “Enhanced” bit in POET’s title implies, this isn’t the first model of its kind — Uber researchers detailed the original POET in a paper published in early January of last year. But the coauthors of this new study point out that POET was unable to demonstrate its creative potential because of limitations in the algorithm and a lack of universal progress measure. That is to say, the means for measuring POET’s progress was domain-specific, meaning that it needed to be redesigned to apply POET to new domains.

Uber AI Enhanced POET

Above: A POET-directed agent navigating an environment.

Enhanced POET has no such limitation, opening the doors to its application across almost any domain.

“Enhanced POET itself seems prepared to push onward as long as there is ground left to discover. The algorithm is arguably unbounded. If we can conceive a domain without bounds, or at least with bounds beyond our conception, we may now have the possibility to see something far beyond our imagination borne out of computation alone,” wrote the paper’s coauthors. “That is the exciting promise of open-endedness.”

As with POET, Enhanced POET takes a page from natural evolution in that it creates problems (e.g., challenges, environments, and learning opportunities) and their solutions in an ongoing process. New discoveries extrapolate from their predecessors with no endpoint in mind, creating learning opportunities across “expanding and sometimes circuitous stepping stones.”

Enhanced POET grows and maintains a population of environment-agent pairs, where each AI agent is optimized to solve its paired environment. POET typically starts with an easy environment and a randomly generated agent before creating new environments and searching for their solutions:

  1. POET generates environments by applying random perturbations to the encoding of environments (numerical sequences mapped to instances of environments) whose agents have exhibited sufficient performance. Once generated, the environments are filtered by a criterion that ensures they’re neither too hard nor too easy for the existing agents in the population. From those that meet this criterion, only the most novel are added to the population. Finally, when the population size reaches a preset threshold, adding a new environment results also in moving the oldest active one from the population into an inactive archive. (The archived environments are used to calculate the novelty of new candidate environments so that previously existing environments aren’t discovered repeatedly.)
  2. POET continually optimizes every agent within its environment using a reinforcement learning evolution strategies algorithm.
  3. After a certain number of iterations, POET tests whether a copy of any agent should be transferred from one environment to another within the population to replace the target environment’s paired agent, if the transferred agent either immediately or after one optimization step outperforms the incumbent.

The original POET leveraged environmental characterizations — descriptions of environments’ attributes — to encourage novel environment generation. But these were derived from hand-coded features tied directly to domains. By contrast, Enhanced POET uses a characterization that’s grounded by how all agents in the population and archive perform in that environment. The researchers say the key insight is that a newly generated environment is likely to pose a qualitatively new kind of challenge. For example, the emergence in a video game of a landscape with stumps may induce a new ordering on agents, because agents with different walking gaits may differ in their ability to step over the obstacles.

Above: A tree of the first 100 environments of a POET run; each node contains a landscape picture depicting a unique environment. The circular or square shape of a node indicates that the environment is in the active population or the archive, respectively, while the color of the border of each node suggests its time of creation: darker color means being created later in the process. The red arrows label successful transfers during a single transfer iteration.

Enhanced POET’s new environmental characterization evaluates active and archived agents and stores their raw scores in a mathematical object known as a vector. Each score in the vector is clipped between a lower bound and an upper bound to eliminate scores too low (indicating the outright failure of an agent) or too high (indicating that the agent is already competent). The scores are then replaced with rankings and normalized, after which Enhanced POET attempts to replace an incumbent agent with another agent in the population that performs better, enabling innovations from solutions for one environment to aid progress in other environments.

Compared with the original POET, Enhanced POET adopts a more expressive environment encoding that captures details with high granularity and precision. Using a compositional pattern-producing network, a class of AI model that takes as input geometric coordinates and when queried generate a geometric pattern, Enhanced POET can synthesize increasingly complex environment landscapes in virtually any resolution or size.

To measure universal progress toward goals, Enhanced POET tracks the accumulated number of novel environments created and solved. To be counted, an environment must pass the minimal criterion measured against all the agents generated over the entire current run so far, and it must be eventually solved by the system so that the system doesn’t receive credit for producing unsolvable challenges.

Uber AI Enhanced POET

In experiments, the contributing team evaluated Enhanced POET in a domain adapted from a 2D walking environment based on the Bipedal Walker Hardcore environment in OpenAI Gym, San Francisco startup OpenAI’s toolkit for benchmarking reinforcement learning algorithms. They tasked 40 walking agents across 40 environments with navigating obstacle courses from left to right, with runs taking 60,000 POET iterations in 12 days on 750 processor cores using Fiber, a distributed computing library in Python that parallelizes workloads over any numbers of cores.

The researchers report that Enhanced POET created and solved 175 novel environments compared with the original POET’s roughly 85 — an order of magnitude leap. The agents improved more slowly after 30,000 iterations, but the team attributes this to the fact that the environments became increasingly difficult from this point and thus required more time to optimize.

“If you had a system that was searching for architectures, creating better and better learning algorithms, and automatically creating its own learning challenges and solving them and then going on to harder challenges … [If you] put those three pillars together … you have what I call an ‘AI-generating algorithm.’ That’s an alternative path to AGI that I think will ultimately be faster,” Clune told VentureBeat in a previous interview.

Source: Read Full Article


DirectX 12 Ultimate unifies ray tracing, speed-boosting graphics tricks across PCs and Xbox

Microsoft has announced DirectX 12 Ultimate, a new version of the graphics technology underpinning both Windows and the upcoming Xbox Series X. Unveiled on Thursday, it’s one of many announcements originally scheduled for GDC 2020 that are carrying on despite the show’s cancellation.

Nvidia shared details of what to expect before Microsoft’s official presentation, and it’s easy to see why: Even though Microsoft’s next-gen console is powered by AMD, DirectX 12 Ultimate enshrines several innovative technologies first introduced by GeForce RTX 20-series graphics cards as a new industry standard—one that now spans both PCs and consoles, earning it the “Ultimate” name.

“By unifying the graphics platform across PC and Xbox Series X, DX12 Ultimate serves as a force multiplier for the entire gaming ecosystem,” Microsoft’s announcement post says. “No longer do the cycles operate independently! Instead, they now combine synergistically: when Xbox Series X releases, there will already be many millions of DX12 Ultimate PC graphics cards in the world with the same feature set, catalyzing a rapid adoption of new features, and when Xbox Series X brings a wave of new console gamers, PC will likewise benefit from this vast surge of new DX12 Ultimate capable hardware! The result? An adrenaline shot to new feature adoption, groundbreaking graphics in the hands of gamers more quickly than ever before.”

DirectX 12 Ultimate supports DirectX Raytracing (DXR) tier 1.1, which is an incremental update to the first iteration. The cutting-edge lighting technology stole the spotlight in GeForce RTX 20-series GPUs to the extent that Nvidia ditched its traditional “GTX” branding for “RTX,” and it’s a key feature of the next-gen Xbox Series X. (The PlayStation 5 also supports hardware-based ray tracing, but Sony’s systems don’t rely on DirectX technology.) The most notable addition in DXR 1.1 is inline ray tracing, a new technique that “gives developers the option to drive more of the raytracing process, as opposed to handling work scheduling entirely to the system,” per Microsoft.

Ray tracing can look positively breathtaking when used to good effect, as it has been in Control and Metro Exodus.  While its adoption has been limited to this point, expect to see ray tracing explode in popularity once it’s in the consoles.

AMD’s forthcoming RDNA2-based Radeon graphics cards will also support hardware ray tracing and DirectX 12 Ultimate, the company has confirmed, along with releasing the video above. 

Ray tracing isn’t the only tech introduced by Nvidia’s Turing architecture that’s being codified by DirectX 12 Ultimate, though. Two other intelligent rendering capabilities make your GPU work smarter, not harder, supercharging the performance potential of your graphics card (or Xbox) if developers wind up embracing them.

DX12 Ultimate hardware also needs to support Variable Rate Shading (VRS) tier 2. Here’s how we described Variable Rate Shading in our Nvidia Turing GPU deep dive:

“Variable Rate Shading is sort of like a supercharged version of the multi-resolution shading that Nvidia’s supported for years now. Human eyes only see the focal points of what’s in their vision at full detail; objects at the periphery or in motion aren’t as sharp. Variable rate shading takes advantage of that to shade primary objects at full resolution, but secondary objects at a lower rate, which can improve performance.

One potential use case for this is Motion Adaptive Shading, where non-critical parts of a moving scene are rendered with less detail. The image above shows how it could be handled in Forza Horizon. Traditionally, every part of the screen would be rendered at full detail, but with Motion Adaptive Shading, only the blue sections of the scene get such lofty treatment.”

Variable Rate Shading can also be used in other ways, such as the Content Adaptive Shading technology that shipped in Wolfenstein II. Content Adaptive Shading applies the same basic principles as Motion Adaptive Shading, but it dynamically identifies portions of the screen that have low detail or large swathes of similar colors, and shades those at lower detail—and more so when you’re in motion—to increase overall performance with minimal loss of perceptible visual quality.

As a high frame rate connoisseur, I’ve been a massive fan of smart rendering techniques like these every time I’ve seen them in action. Fingers crossed that with VRS now part of DX12 Ultimate and the next-gen consoles, developers will start deploying tools like these more often.

Microsoft’s latest version of the API also demands support for mesh shaders, per Nvidia. Once again, Nvidia’s Turing GPUs introduced the concept of mesh shading. Here’s what we said at the time:

“Mesh shading help take some of the burden off your CPU during very visually complex scenes, with tens or hundreds of thousands of objects. It consists of two new shader stages. Task shaders perform object culling to determine which elements of a scene need to be rendered. Once that’s decided, Mesh Shaders determine the level of detail at which the visible objects should be rendered. Ones that are farther away need a much lower level of detail, while closer objects need to look as sharp as possible.

Nvidia showed off mesh shading with an impressive, playable demo where you flew a spaceship through a massive field of 300,000 asteroids. The demo ran around 50 frames per second despite that gargantuan object count because mesh shading reduced the number of drawn triangles at any given point down to around 13,000, from a maximum of 3 trillion potential drawn triangles.”

It was super impressive in the demo. You can see Nvidia’s asteroids demo in action below, and read more about how it works here. No games have taken advantage of mesh shading thus far, even though GeForce RTX 20-series GPUs have been available for well over a year. Nvidia expects to see traction for the technology now that it’s part of DX12 Ultimate.

If you want to learn more about how VRS can be implemented, check out the GeForce blog post on Nvidia Adaptive Shading in Wolfenstein: Youngblood. Youngblood’s NAS combined two different forms of Variable Rate Shading—Content Adaptive Shading and Motion Adaptive Shading—to “accelerate performance by up to 15%, with no perceivable visual quality loss.”

Finally, DirectX 12 Ultimate also requires support for a technology called Sampler Feedback. Microsoft says that Sampler Feedback “enables better visual quality, shorter load time, and less stuttering.”

“Fundamentally, what this does is allows the texture subsystem to provide some communication to the shading subsystem, to allow developers to make more intelligent choices about the way it manages sampling, filtering, and LODs (levels of detail),” Nvidia’s Tony Tamasi explained during a press briefing. “The way we demonstrated that is with texture space shading. The idea there is that certain shading can be reused, either spatially or temporally across space or time, to allow developers to potentially save performance that don’t need to be reshaded versus shaded time and time again when it’s not necessary.”

For VR workloads, Tamasi said you can use Sampler Feedback to render all the pixels in a scene once, then simply reproject it to the second eye, “saving essentially half the shading calculationse.”

“You can do the same kind of thing temporally,” Tamasi continued. “Say you have a mountain off in the distance, and you didn’t have a lot of dynamic lighting shifting across that mountain. You could potentially update the shading on those mountains every other frame, or every tenth frame, whatever you felt was appropriate for your game, again saving all those shading calculations, allowing developers to up the image quality by putting the quality where it matters, where users are going to see it, and saving some performance in places where it’s not really going to matter.”

Fascinating stuff.

GeForce RTX graphics cards like the RTX 2080 Super already support DirectX 12 Ultimate.

Beyond the fact that many of these technologies debuted in Nvidia’s “Turing” GeForce graphics cards well over a year ago, it’s also worth noting that aside from ray tracing, the other DirectX 12 Ultimate capabilities focus on having GPUs work smarter to increase performance. Real time ray tracing puts big-time stress on your graphics chip even with hardware dedicated to the task. Augmenting it with software tricks that can boost performance (including Nvidia’s DLSS on PCs) isn’t just wise, it’s borderline necessary.

Any PC hardware needs to support all of these technologies to be DirectX 12 Ultimate compliant. Nvidia’s GeForce RTX graphics cards already do, obviously, as will the Xbox Series X and AMD’s next-gen RDNA2 Radeon graphics cards scheduled to launch later this year.

For a much more technical deep dive into DirectX 12 Ultimate’s capabilities, be sure to check out the post from Microsoft’s DirectX team.

Editor’s note: This article was originally published on 3/19/20, but was updated to include information from Microsoft’s DX12 Ultimate announcement post.

Source: Read Full Article


Diligent Robotics raises $10 million for nurse assistant robot Moxi

Diligent Robotics today announced the close of a $10 million Series A round to expand its fleet of nurse assistant robots for hospitals.

Moxi is made to help reduce nurse workloads by doing things like collecting supplies, gathering soiled linens or delivering fresh ones, and comes to the market  at a time when there’s a nurses of shortages. Help can also mean less exposure to disease for health care professionals sorely needed during the coronavirus pandemic.

Moxi was created by Diligent Robotics at University of Texas, Austin by CEO Dr. Andrea Thomaz, a roboticist and professor who previously ran the Georgia Tech Socially Intelligent Machines Lab.

“It’s a really good time to be working on this problem,” Thomaz told VentureBeat in a phone interview. “Now more than ever there’s kind of a spotlight on how overworked and what a shortage there is of frontline hospital staff, so we’re anxious to get more robots out there to help.”

Robots like UVD’s ultraviolent light-emitting robot and others used for telepresence are being deployed to fight coronavirus.

Moxi carried out more than 3,000 hours of testing in Austin and Dallas, Texas hospitals in 2018 and 2019, and earlier this year started work with its first customer, Medical City Healthcare in Dallas, Texas. The company has not yet shared specific data on just how much time Moxi gives back to human nurses, but those figures may be shared in a future white paper, Thomaz said.

Diligent Robotics previously raised $5 million in a seed round. The $10 million round was led DNX Ventures with participation from True Ventures, Ubiquity Ventures, Next Coast Ventures, Grit Ventures, E14 Fund, and Promus Ventures. Diligent Robotics was founded in 2017, has 18 employees, and is based in Austin.

Source: Read Full Article