Digital twin firm Duality AI launches cloud-based platform for testing robots – SiliconANGLE

Duality AI, the developer of a suite of tools that allow for simulating the real world so artificial intelligence and robots can navigate it using “digital twins,” announced today that it’s expanding its capabilities into the cloud, making it possible to access simulations through web browsers.

Duality develops a product known as Falcon, which is a powerful 3D virtual world product built on the Epic Games Inc.’s Unreal Engine. It’s capable of simulating real-world conditions and recreating objects such as terrain, buildings, cars, furniture and other obstacles otherwise known as digital twins.

With FalconSim, the developers of AI-enabled robots and self-driving vehicles can test out their creations safely in virtual worlds to see how they will act and react in simulation before deploying them in the real world. Now with the launch of FalconCloud, that capability has been moved from desktops and onto web browsers. That will allow customers to run simulations at massive scale without the need for special hardware and software to test and simulate AI-powered robot interactions.

Duality was co-founded by Chief Executive Apurva Shah, a veteran of the entertainment industry who worked for Pixar Animation Studios making virtual spaces such as Paris for the movie “Ratatouille,” and Chief Product Officer Michael Taylor, an engineering and autonomics robotics expert who worked for equipment manufacturer Caterpillar Inc. This complementary mix of talents, Shah told SiliconANGLE in an interview, led the pair to combine their efforts and build the company that exists today.

“We thought, can we train robots and AI in these virtual worlds? And that’s how we got started,” Shah said. “It’s an easy problem to describe, but then getting the data that actually works for an engineering context that’s repeatable, deterministic and high fidelity enough to train AI models. That’s obviously a more difficult problem.”

Shah said that Duality has spent the past few years working on this problem and that it has been paying off. The simulation capabilities, which he calls “embodied AI,” have been put to use by customers such as the United States Defense Advanced Research Project Agency, an organization tasked with the development and research of emerging technologies for use by the military, multinational technology conglomerate Honeywell International Inc. and aerial autonomous drone developer KEF Robotics and Procter & Gamble Co.

“A lot of what motivates us at Duality is we believe that these smart systems are coming to the question is: How do you make them really safe? And safe, not just when it’s nice and sunny, but safe when conditions are not optimal,” said Shah. “How well does an image-based sensing system work when there is heavy fog or dust cloud, right?”

The first death attributed to a self-driving car happened in 2018 when an autonomous Uber collided with a pedestrian at night in Tempe, Arizona. The vehicle detected the woman crossing the road, but failed to identify her as a person and struck her.

Using Falcon’s technology, AI robot systems and their sensor packages can be simulated in different situations, including darkness, rain, sleet, snow, fog and other inclement conditions. They can also be given different obstacles or even various situations, such as irregular traffic patterns or people wandering into their path.

Along with FalconCloud, Duality also announced FalconEditor, which allows users to quickly and easily build their own simulations using Unreal Engine 5. This rendering engine contains one of the largest sets of 3D model assets and level editors in the world, along with physics simulations that can be used to recreate diverse real-world simulations and scenarios using digital twins.

“Then other team members can very quickly and easily run these scenarios and generate data,” Shah said. “Ultimately, if you’re an AI engineer or robotics engineer, you are interested in using the simulations to get to the data that you need. So we’ve tried to make that very, very easy for the whole team to collaborate around.”

Additionally, FalconSim received numerous updates that improved its own software-based virtual sensors, including enhanced sensor rendering capabilities that better match real-world sensor types.

“In the past, KEF often faced a tradeoff between simulating expansive environments and maintaining high fidelity,” said Fraser Kitchell, chief executive of KEF Robotics. “With Duality’s Falcon simulator, this compromise is no longer a concern. Falcon enables us to rigorously test our autonomy algorithms in photorealistic environments an order of magnitude larger than any we’ve previously used.”

Shah said he sees robotics, especially autonomous AI-powered smart systems, proliferating quickly right now and Duality will be following its customers’ journeys by giving them the tools to find their way safely. This means learning to get comfortable doing their job by providing them a way to test their systems in a virtual world with digital twins first at the design and development stage before they deploy into the real world, which can save them a lot of time and money.

Image: Duality AI