After testing its concept networking technique, the Bonsai found its artificial intelligence platform can enable robots to perform tasks at 45 times the speed of comparable machine learning techniques, the company announced.
The concept networking approach speeds up the training process by breaking tasks down into multiple sub-tasks, each of which the machine learns to do separately through a variety of learning techniques. After learning each sub-task, the robot combines them to perform the intended action.
“I see a wide variety of problems this could solve,” CEO and founder Mark Hammond said in a Forbes article. “This is the cutting edge of applying AI to controller and optimization problems.”
Here’s how the research team applied this idea to train a robotic arm to pick up and stack a block. The action was broken down into five tasks: reach, orient, grasp, move and stack. The robot already knew how to complete the tasks reach and move, and so didn’t need to relearn these. However, it had to be trained in the tasks of orient, grasp and stack through trial-and-error, a process called reinforcement learning. When all five tasks were mastered, the robot arm combined them to pick up and stack the block.
This concept builds on the reinforcement learning concept employed by Google’s DeepMind. However, because the concept networking approach doesn’t require a machine to attempt a full action over and over again in a simulated environment until it accomplishes it, Bonsai’s approach ends up being more efficient — 45 times more to be exact.
“Building off the foundation established by DeepMind, we were able to achieve these results by combining state-of-the-art reinforcement learning techniques with innovative features that are unique to the Bonsai Platform,” head of AI Marcos Campos said in a press release. “Using Bonsai, enterprises now have access to the tools and technology to program control systems more efficiently than any other commercially available reinforcement learning platform.”
The development of AI platforms moves forward the creation of autonomous machines, both industrial and commercial. The platforms — including those from Bonsai or DeepMind or even the Enable Mobile Machine Automation navigation system from Brain Corp — are needed to act as the machines’ brains to allow them to perform their designated tasks without human direction.