Automated Design of Vibratory Part Feeding Traps for Controlled Reorientation of Parts

One of the most challenging tasks in automating the design of assembly systems is to automate the feeding process, which consists of singulating parts arriving in bulk, recovering their poses, and presenting them a controlled manner. Vibratory part feeders are the most prevalent feeder type in industry due to their speed and reliability. Recent work has shown great promise in automating the design of vibratory feeders which in a structured manner output parts in one of their natural resting poses. However, robotic assembly operations often require parts to be grasped in a specific orientation for subsequent manipulation. To avoid the need for regrasping prior to assembly, it is essential that the set of poses that parts can be presented in is not restricted to the set of natural resting poses. We present a method for automatically generating trap designs that can rotate parts around the axis of propagation in a controlled manner. The design relies on dynamic simulation and Wilson Score Kernel Density optimization to automatically find suitable trap parameters. We evaluate the trap designs in simulation and demonstrate the desired real world behavior by 3D printing optimized trap designs and testing them on a linear vibratory feeder.