Northwestern University’s Noodlebot is designed for future applications. (Image Credit: Northwestern University)
Smart robotics are about to get more reliable thanks to a new AI algorithm called Maximum Diffusion Reinforcement Learning (MaxDiff RL) developed by Northwestern University engineers. Robots with the MaxDiff RL algorithm can learn new experiences by observing the surrounding environment randomly. More impressively, robots learn and execute new tasks on the first try via rapid task acquisition. During experiments, the team realized this algorithm, when used with simulated robots, outperformed other advanced AI models.
Machine-learning algorithms are typically trained via curated big data. AI then learns from that training until it achieves desirable results. According to Northwestern University engineers, robots won’t be able to learn through this technique even though it works for ChatGPT and Google Gemini. Robots autonomously collect data, so they don’t need human curators.
Conventional algorithms aren’t compatible with robots in two ways. “First, disembodied systems can take advantage of a world where physical laws do not apply. Second, individual failures have no consequences,” said Todd Murphey, a professor of mechanical engineering at McCormick.“For computer science applications, the only thing that matters is that it succeeds most of the time. In robotics, one failure could be catastrophic.”
So, the team developed a novel algorithm for robots to gather high-quality data on the go. It instructs them to move more randomly for thorough, diverse data collection about their environments. Relying on self-curated random experiences allows a robot to gain the essential skills to perform certain tasks.
The team tested their new algorithm against advanced models. They used simulations to instruct the simulated robots to perform a set of tasks. As a result, robots that used the MaxDiff RL acquired knowledge faster than other models. Additionally, they proved more consistent and reliable in comparison. And these robots didn’t need multiple attempts to succeed --- it only took one try, even without prior knowledge.
MaxDiff RL is also designed to handle varying applications. The team hopes it can solve issues that hold back the field, leading to reliable decision-making in smart robotics. “This doesn’t have to be used only for robotic vehicles that move around,” Ph.D candidate Allison Pinosky said. “It also could be used for stationary robots — such as a robotic arm in a kitchen that learns how to load the dishwasher. As tasks and physical environments become more complicated, the role of embodiment becomes even more crucial to consider during the learning process. This is an important step toward real systems that do more complicated, more interesting tasks.”
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell