Microsoft
Project Moab
A playground for mastering intelligent control systems
The Challenge
Microsoft’s Project Bonsai solution enables engineers without a background in data science to apply their subject matter expertise and accelerate the development of intelligent control systems. We partnered with Microsoft to design a ball-balancing robot that showcases the power of their AI solution, equipping engineers with the knowledge and expertise needed to solve their novel use cases using Bonsai’s power.
Our Solution
Moab, a ball-balancing robot, is operated using motion control, visualized in simulation on Microsoft Project Bonsai, coordinated using trained Project Bonsai brains, and then deployed to the physical bot. Our team built Project Moab to complete the initial objective of balancing a ball via machine learning. Creating a robot powered by Bonsai required a solution incorporating computer vision, artificial intelligence, and hardware, requiring a cross-disciplinary collaboration between Fresh and Microsoft.
Built with Project Bonsai: a ball-balancing robot
Moab, a ball-balancing robot, is operated using motion control, visualized in simulation on Microsoft Project Bonsai, coordinated using trained Project Bonsai brains, and then deployed to the physical bot.
Our team built Project Moab to complete the initial objective of balancing a ball via machine learning. Engineers can teach Moab to catch balls thrown toward it or even after they bounce on a table. Like a self-contained game of labyrinth, Moab also learns to balance objects while ensuring they don’t come into contact with obstacles placed on the plate.
Creating a robot powered by Bonsai required a solution incorporating computer vision, artificial intelligence, and hardware, requiring a cross-disciplinary collaboration between Fresh and Microsoft.
Democratizing AI development
The Project Bonsai platform makes machine teaching, a paradigm pioneered by Microsoft, accessible to engineers from various backgrounds, even those with minimal AI expertise. With machine teaching, engineers break complex problems into individual skills and give AI brains important information about solving the problem faster.
Pairing machine teaching with elements of reinforcement learning (RL), Microsoft has equipped engineers to create AI that learns by executing decisions and receiving rewards for actions that get it closer to an end goal. While traditional reinforcement learning is a time-consuming, brute force, trial-and-error approach, machine teaching accelerates and improves the training process and even allows engineers to reuse the individual steps for other AI brains in a simulated, auditable environment.
That’s where the power of Moab comes into play: engineers from various backgrounds can realize the potential of Project Bonsai, mastering the skills needed to implement intelligent control systems and exploring the value they hold.
End-to-end product design & development
Creating Moab was a holistic, interdisciplinary effort. From programming the software to designing the hardware, collaboration guided our process. At every stage, the end-user’s experience was considered while keeping in mind business, design, and manufacturing constraints.
Software design
A prototype-first methodology
Moab’s success is determined by how well it learns to balance the ball over time. Creating a Moab prototype allowed the software team to write code and test components, pushing the design forward. Using a prototype-first methodology, we translated abstract ideas and concepts into a testable physical product, where software development efforts supported the ultimate hardware experience.
Creating a custom PCB
Printed circuit boards, or PCBs, often come as off-the-shelf solutions. For Moab, this wasn’t possible given the unique requirements and specifications of the project. Our engineers and industrial designers collaborated to create a custom PCB with all of the extensive inputs necessary while integrating seamlessly into the robot’s unique size and shape. A custom PCB also allowed us to meet special requirements and specifications while maintaining reasonable cost, with the eventual goal of creating hundreds of Moab units.
Ideation & iteration
Our approach to industrial design
A prototype-first methodology was also essential for industrial design. Given the cross-section of skills and disciplines required to create the Moab bot, the ID team designed and built numerous variations to learn what worked, what didn’t, and why, which ultimately advanced the design effort.
A manufacturing collaboration from across the world
Working with an experienced overseas manufacturer streamlined our process. We initially used the process of injection molding, shooting Moab’s metal mold with liquid hot plastic that cooled, forming the part. This allowed our team to see through the plastic to test fitting and alignment, confirming that the specifications were exact before getting them into the correct color, material, and finish.
Considering CX
Creating a complete package
The packaging is one of the first experiences a customer has with a product. We wanted Moab’s simplicity, clarity, and sophistication to reflect what customers first see when it arrives.
We opted for a minimalist packaging concept, from the words printed on the box to how it unfolds. The result is a product that, from packaging to performance, helps engineers learn to develop intelligent control systems and communicates the value of human-centered design thinking.
2022 Red Dot Award
Winner of “Best of the Best” in 2022 Red Dot Design Awards
