Vision based robotic portion control for granular food

The food industry is increasingly adopting robotic automation for food preparation, driven by the need for precision and efficiency. Vision systems play a pivotal role in enabling robots to execute tasks effectively in these dynamic and complex environments. This work proposes two vision-based methods for real-time estimation of the mass of food collected by a robotic manipulator, allowing precise portion control for granular foods, such as rice. Specifically, we investigate the usage of 2D and 3D vision algorithms. The former estimates the quantity of food based on the area of the collection tool that is occluded by food, while the latter obtains the estimate from the volume of food collected. Experimental results demonstrate that the 2D algorithm excels at higher speed, while the 3D algorithm offers superior accuracy at lower speed, highlighting the strengths and limitations of each approach for different operating conditions.