Product reviews, deals and the latest tech news

Robots find it surprisingly difficult to pick up a chicken wing, as demonstrated by Nvidia

It’s the American football playoffs, a time when mindless snacking on chicken wings is all too common. A lot of work went into picking up that piece of chicken by the robot that helped make your meal.

On Thursday, Nvidia demonstrated how the Massachusetts-based firm Soft Robotics is using its technologies, such as local GPUs and the Isaac Sim robotics simulation tools, to simplify the rollout of robots suited to handle meals like chicken wings.

It could make sense to put robots to work in factories that process and package food. Rapid movement of conveyor belts transports precooked, consistently prepared foods like chicken wings to waiting customers. Meanwhile, businesses like the meat processing industry are struggling to find workers willing to take on the additional health and safety hazards associated with these positions.

Nevertheless, the food business has been hesitant to embrace robotic equipment. Robots have struggled for a long time to successfully grasp items with unconventional shapes. A piece of chicken, for example, might be especially difficult to grasp because of its slick and glossy surface. In addition, any food that is lost because a robot knocked it to the floor or dropped it is wasted.

Computer vision and training on data sets that may teach the system to recognise a single item of food to grasp are essential for a robotic system to function on the food processing line in addition to the necessary hardware. Large food manufacturers like Tyson Foods and Johnsonville rely on these technologies, which Soft Robotics offers. The company’s mGripAI system can be taught to pick up and handle a wide variety of meals, including meats, veggies, and baked products, using separate training datasets. Tyson Ventures, Marel, and Johnsonville Ventures provided the firm with $26 million in Series C capital.

Nvidia claims that the Omniverse platform has helped Soft Robotics reduce the time it takes to instal its mGripAI from months to days. In particular, Isaac Sim, a platform for creating digital twins of manipulation robots, is being used by the firm. Nvidia’s Omniverse Replicator, a synthetic data-generation engine, served as the foundation for Isaac Sim, which was released in 2021. Because having access to high-quality data is crucial for developing AI models, Nvidia created these tools to address a common problem faced by a large proportion of AI projects: a lack of it. Data from Replicator may be used to construct robots that can learn new abilities in a variety of simulations before applying them in the real world.

Soft Robotics utilises Isaac Sim to generate 3D images of chicken pieces in a variety of locations (conveyor belts, bins, etc.) and lighting conditions in order to train AI systems for chicken producers like Tyson Foods. The simulations teach the AI system how the chicken may appear if it were piled and how to best collect the parts it needs.

As a participant in the Nvidia Inception programme, which offers businesses GPU assistance and coaching on AI platforms, Soft Robotics is also leveraging on-premise GPUs.