Our software and AI models are game changers, allowing us to train models to adapt to variability so that one robot can pick and plate any ingredient efficiently and accurately into any container 🤖
Chef uses AI to discern where to pick ingredients from and where to place them. The more data our AI models ingest, the better their performance becomes. This video is a good visual demonstration of how our software works! On a high level - To pick, our models take into consideration the AI policy for that ingredient class, the topography of the food it's picking from, any occlusions and a memory map of past picks, as well as the utensil the robot is using (and the correlated expected sweep volume from picking) to figure out the optimal pick pose. After it picks, Chef will use a 1 gram resolution scale and some signal processing to figure out how much weight was actually picked. Once we have this, we'll see how close our AI policy was to the target weight; based on this, it'll adjust how it picks the next pick. - To place, another model will detect and track the trays (along with estimating speed) as they move down the line. This model will react to things like changes in tray orientation, distances of the tray from the robot, skew in the conveyor from the robot, slope of the conveyor, conveyor stops, changes in conveyor velocities and accelerations, and any last minute additions or removals of trays that can affect tray tracks. All of this allows Chef to place food into a tray the way the customers wants - e.g., into a particular compartment, spread over the tray, or in a particular part of the tray while minimizing spillage.