The most malleable medium
Software is the most malleable medium that there is. This malleability enables us to craft abstractions and layers of capability that allow creators to quickly build any type of autonomous systems that can be imagined.
Many early efforts to create autonomous systems were developed as siloed, bespoke applications by a team of experts. We knew we needed a different approach, so we looked to provide open, reusable and powerful tools and a platform that would enable anyone to teach intelligence to machines without being an AI expert. Engineers, experts in their own fields, can transfer their years of experience into an AI solution without going back to college to learn AI.
But how do you teach a machine? Most machines today are hardcoded in functionality and have basic or rigid control systems. Our tools allow an expert to express what needs to be learned and how to learn it. With that input, our platform uses reinforcement learning against a simulated environment to build the AI solution. The resulting AI solution can now control the machine and make it adapt intelligently to the real world, in the same way a human would. For example, instead of building special cages for robots, the robots learn to climb stairs, open doors, navigate a dynamic warehouse, land a drone or work together with other systems and humans.
The Bell team identified Microsoft technologies that could help them work toward their goal of transforming their business model from helicopters to autonomous flight vehicles. Along with computer vision and cloud computing from Azure, Project Bonsai and AirSim helped Bell create an AI solution that can practice in a simulated environment.
Matt Holvey, Senior Manager of Intelligent Systems at Bell, had this to say: “We’re using Project Bonsai because it allows us to quickly create and teach an AI just like if we were training a pilot on what to look for. You can get the AI to understand what decisions to make about altitude and pitch based on the identified landing zone it sees.”
Machine teaching enables any engineer to safely add autonomy to new or existing machines.