Paul McLaughlin

MADEit was the most ambitious project I’ve been involved in. The vision was to revolutionize manufacturing by creating reconfigurable manufacturing cells. These cells could dynamically adapt their tooling, inventory, and processes, forming a modular system capable of manufacturing a vast assortment of products quickly and efficiently. The idea was to turn manufacturing into a software-driven problem, allowing for rapid iterations and seamless technology adoption.

The upsides were enormous. It would remove the need for traditional manufacturing infrastructure, drastically lowering barriers to entry for new products. New technologies could be integrated on the fly, allowing products to improve continuously. Local manufacturing would reduce shipping times and costs, while zero-inventory, on-demand production would make customization essentially free. However, the complexity of this audacious project was greater than anticipated.

Initially, we chose to use a robot arm as our solution. Robot arms are versatile and could be adapted for various tasks. We saw this as a long-term investment, not only for cooking but also for future manufacturing processes. However, this design quickly proved too complex for the task at hand. The robot arm had too many moving parts, reducing its reliability and increasing the cost and maintenance.

Recognizing the need for simplification, we moved away from the robot arm and began exploring other designs. Our first step was prototyping a cable-driven robot, which used three motors and cables to position the tool in 3D space, guided by a laser system. This approach was effective in theory, but in practice, it was too finicky for 24/7 operation. The cables required constant tension adjustments, and the system wasn’t robust enough for the reliability we needed.

Next, we shifted to a delta robot. Delta robots are known for their simplicity, ease of design, and speed. We were able to build and program the first prototype in just a couple of weeks. While this design worked well for certain tasks, it had limitations when it came to grasping and manipulating items from the inventory. The physical constraints of the delta configuration made it unsuitable for the diverse handling needs in the kitchen.

Finally, we settled on a gantry robot, which turned out to be the ideal solution for our needs. Gantry robots operate on a linear axis, providing full access to the inventory while maintaining a simple design that is both easy to build and maintain. Its simplicity was a significant advantage, and the cost was far lower than other options we had considered. Looking back, we only wish we had realized this earlier in the process(!!)

While the robot was the key component of the system, it was just one piece of a larger puzzle. Our manufacturing cell included several critical elements: inventory cells for storing food, a robot designed to move the inventory through the system, specialized tooling for cooking, and a resource cell providing the necessary cooling, compute power, electricity, and connectivity to keep everything running. At the heart of it all was our software, which orchestrated the entire process, ensuring that each component worked together seamlessly. The video below demonstrates how the system functions, from receiving an order to cooking and delivering a meal efficiently.

The final vision for MADEit was a standalone robotic kitchen housed in a mobile trailer, capable of being deployed anywhere. The concept was to deploy multiple units in parking lots, connected directly to existing delivery services. These robotic kitchens would be able to operate autonomously, cooking food on-demand for local delivery without human intervention.

The final step was determining how to keep these autonomous kitchens supplied, so we quickly mocked up an efficient distribution center to support them. Overall, the project was incredibly challenging, especially as I was new to robotics, which meant a steep learning curve in designing, programming, and building them. Hardware development posed its own unique difficulties; parts would often take weeks to arrive, and each iteration took days.

Eventually, we streamlined the process, but the phrase "hardware is hard" proved true on many levels. Now, with a focus on writing AI software, the speed and flexibility of updates and adding new libraries is something I do not take for granted.