More variation, less volume: simulations that allow robots to be deployed, reprogrammed and scaled
While the field of robotics is advancing fast, what’s needed most to accelerate its development, according to Amit Goel, is more humans.
The director of product management – Edge AI and Robotics at intelligent computing giant Nvidia adds that there’s a lot more work needed to move the industry forward.
“The tools need to be easier to use so that developers are not intimidated. It should be that you don’t need a PhD before building a robot. A lot of entrepreneurial innovation has to happen for the robot to become more mainstream,” he says.
He compares the number of developers today working on apps for mobile phones to those working on robotics. “The numbers are way skewed to the app.”
At the same time there is rising demand for robotics to plug the widening gap in labour shortages. This issue is particularly acute in supply chain logistics and exacerbated during the pandemic when workers were either locked out of warehouses or fell sick.
It’s also a key reason why the global robotics market is projected to grow at over a 25% rate annually, an increase from approximately 20% prior to Covid, to reach $210 billion by 2025, according to Statista.
“The initial robotics market was predominantly in large scale manufacturing and is now being adopted by small and medium scale businesses,” informs Goel.
“In automotive manufacture, robotics were used for doing repetitive tasks in large volume with little variance but the move into smaller businesses requires more variation and less volume. That means robots need to be easy to program, deploy, reprogram and scale.”
That’s where advances in artificial intelligence (AI), semiconductor technology, mobile edge computing and more affordable sensor capability come in. Combined, they are helping to bring robotics and automation, previously considered out of reach of many markets to users in everything from healthcare to agriculture and food delivery.
Next gen bot plans
Nvidia made $5bn in revenue in 2021 mainly from sales of microprocessors but its software products are strong too. Both hardware and software divisions are targeting robotics. The goal of its Robotics Research lab based in Seattle is to develop the next generation of robots that can robustly manipulate the physical world and work alongside humans.
“Across the board we see huge growth in robotics and automation fuelled by the fact that there’s a lot more ecommerce and packages moving through the system,” Goel says. “Also, so many baby boomers are retiring right now and leaving many jobs unfilled which is why automation has become critical for some industries.”
For that reason, he says, robots in the enterprise are not so much eliminating jobs but improving overall productivity.
“The new wave of robots are enabling things not possible before. Think about an assembly line. Before, everything was fixed so you could only manufacture things in a certain order. Now with AMRs (Autonomous Mobile Robots) you can take a part anywhere you want, you can change the layout anytime. It enables just-in-time time configuration that allows you to get more efficiency out of your factory.”
Using AMRs at industrial scale, even small routing optimisations could save billions of dollars in the $9 trillion logistics industry. Research firm ABI estimates that the market for industrial robots will double from 406,000 units in 2021 to 815,000 in 2030, reaching $27bn in value.
Many AMRs in logistics and warehouses “don’t need that .01mm precision” of more advanced applications, so the cost of the tech is more affordable.
The lifecycle of this new generation of AI robots includes some common pieces of technology. Since all robots run on data the first port of call is to collect data for the specific application. Nvidia offers Replicator that enables synthetic data generation.
“Many industries are not equipped with sufficient sensors and do not have a lot of historic data to use for training the AI model,” Goel says. “We develop the tools and platforms that allows you to create large amounts of data and we provide the entire AI framework to train AI models in the cloud or at the edge all the way onto the robot.”
The Isaac sim
Since robotics development teams tend to be spread across the world, Nvidia’s Omniverse platform – built for enterprise 3D design collaboration and digital twin simulation – enables them to work in an online environment capable among other things of simulating real world physics.
The Isaac Sim (built on Omniverse) simulates the behaviour of robot fleets, people, and other machines using digital twins with high-fidelity physics and perception. It assists in robotics design, verification of sensors, validating performance of the algorithm and team collaboration.
“There have been robot sims before but being able to collaborate within the simulation environment is game changing,” Goel says.
Once trained in a sim you need to deploy intelligence onto the robot itself using High Performance Computing. Nvidia’s AI computer Jetson can process up to 275 Trillion Operations per Second (TOPS) for realtime support of multiple sensors.
“With Jetson we are bringing server class computing in a small form factor so it can live in a robot brain,” he explains. “You can connect it to all your cameras and run computer vision and control.”
Once deployed the robot will have actual data to learn from. Data can be streamed from the robots to a digital twin existing in the Omniverse for it and its behaviour to be continually assessed and moderated, retrained, retested, and redeployed.
Amazon, for example, uses Nvidia Omniverse and Isaac Sim to simulate warehouse design, train robot assistants and gain operational efficiencies before physically implementing them in warehouses. PepsiCo is using Omniverse to optimise warehouse layouts and workflows to accelerate throughput before making physical investments or executing changes.
Nvidia runs its own startup booster program called Inception and aims to reduce the time it takes for new robots to be built. It recently joined forces with Open Robotics to add the widely used open-source ROS 2 (Robot Operating System) to Omniverse. The partnership essentially combines the two most powerful robotics development environments and the two largest groups of robotics developers.
“It’s all very well to programme a single robot but it needs to exist in an ecosystem cohabited by other machines and people. That is where the digital twin and Omniverse is important since we can simulate what happens to a robot in context.”
That’s not to say there aren’t considerable challenges ahead. Although designed to shore up the world’s supply chain, the supply chain to actually make and ship robots remains fragile.
“Customers may be ready to order but there are just not enough parts available,” says Goel. “Another factor is that the level of upscaling needed to operate these robots needs addressing. All robots have a degree of autonomy but they still need to be managed and supervised by people and that requires new training of the workforce.”
In addition to which, the technology itself is still nascent. “We’re just scratching the surface,” says Goel. “If the long term goal is to make seamless robotic deployment then among other things we need better algorithms and more intelligent systems.”
To become more intelligent and therefore useful the robotics industry needs more compute capacity to run more sensors and more algorithms to better perceive and understand the world. The robotics market may reach $210bn in a few years but that is a fraction of the value of the products and services that will eventually be generated by its advance.
As Goel says: “We provide the tools so that your robot can live tens of years of life in a few days. That is how you accelerate development and make it more affordable.”
Subscribe to our Editor's weekly newsletter