AmbiKit
At Ambi Robotics, I designed a robotic system that automates the kitting process of monthly subscription boxes. The goal was to make a system that could group five items together so operators at the end could sweep the items into the final container that is shipped to the customer.
The system used five Yaskawa robot arms which were powered by Ambi OS, a proprietary machine learning model which allowed the robots to pick small items out of unstructured tubs.
This project had two stages and the final five-robot system was installed at the customer’s facility in only ten months.
The first step in the project was building a proof of concept. Both the customer and my team wanted to make sure that this was a product that would really work for them, without designing a full scale five robot system. I designed a single robot cell and we installed it in their facility. The single robot design was an extremely helpful first step in the project because it allowed us to validate that the combination of our neural network and robot end effector could successfully pick the customer’s products as well and allowing me to validate the basic dimensions of the system. We also used this first prototype to confirm that the ergonomics of the output station worked with the operators.
After we validated the proof of concept. I scaled the design up to a five robot line. Each robot picked a single product from the blue bins and placed it onto the cleated conveyor in front of it. This system had over five times the throughput of the initial proof of concept, and met our customer’s demanding throughput goals.
Besides scaling the system up from one to five robots, my main innovation for this final version of the product was the “sweeper.” When the cleated conveyor moved the product out of the robot cells, the operator would press a button which unlocked the sweeper and allowed them to pull the product off of the conveyor into small cubbies. They could then use the cubbies to slide the product into the final output container.