Nvidia shows how surprisingly hard it is for a robot to pick up a chicken wing

It’s football playoff season in America — a time when it’s all too easy to pick up and chow down on a chicken wing without much thought. It turns out, however, that the robot that helped prepare your chicken had to put in a great deal of effort to pick up that piece of meat. 

Nvidia on Thursday showcased how the Massachusetts-based startup Soft Robotics is deploying its technology — including on-site GPUs and the Isaac Sim robotics simulation toolkit — to make it easier to deploy robots designed to handle foods like chicken wings. 

Also: Nvidia’s robot simulator now includes human characters too

Food processing and packaging plants may seem like an obvious place to deploy robots. Foods like chicken wings are quickly moved across conveyor belts as they’re uniformly cooked and prepared for consumption. Meanwhile, employers like meat-packing companies are having a difficult time filling jobs that come with increased health and injury risks. 

Even so, the food industry has been relatively slow to adopt robotic tools. Gripping unusually-shaped objects has long been a major challenge for robots. On top of that, something like a piece of chicken can be slippery and shiny, making it even harder to grab. And if a robot drops a food item or knocks it to the floor, it goes to waste. 

Along with the right hardware, a robotic system on the food processing line requires computer vision and training on data sets that can teach the system how to spot a single piece of food to grab. Soft Robotics provides these kinds of systems to major food suppliers like Tyson Foods and Johnsonville. The startup’s mGripAI system can be trained with unique datasets to pick up and handle different kinds of foods, such as meats, produce or baked goods. The company recently raised $26 million in Series C funding from Tyson Ventures, Marel and Johnsonville Ventures. 

Soft Robotics has managed to speed up its mGripAI deployments from a matter of months to days using Nvidia’s Omniverse platform, Nvidia says. More specifically, the startup is using Isaac Sim — a platform for building digital twins of manipulation robots. Isaac Sim, which Nvidia rolled out in 2021, is based on Nvidia’s Omniverse Replicator, a synthetic data-generation engine. Nvidia built these tools because data is a necessary prerequisite for building AI models — however, a significant number of AI projects suffer from a lack of high-quality data. Robots built using data generated by Replicator can master skills across a range of virtual environments before applying them in the physical world. 

To build AI systems for chicken suppliers like Tyson Foods, Soft Robotics uses Isaac Sim to create 3D renderings of chicken parts in different settings — such as on conveyor belts or in bins — and with different lighting scenarios. The simulations help the AI system understand how individual pieces of chicken might looked stacked up on top of each other and which will be easiest to grab. 

Soft Robotics is also using on-site GPUs, and it’s part of the Nvidia Inception program, which provides companies with GPU support and AI platforms guidance.