Computer scientists and robot cists frequently draw inspiration from animals and other live organisms while developing new technology. This enables them to simulate complicated behaviors and movement patterns in order to improve the performance, efficiency, and capacities of their systems.
Researchers from Zhejiang Sci-Tech University and the University of Essex recently created a reinforcement learning technique for controlling the movements of a single-legged robot inspired by beavers. Their technology, described in a study published in Robotics and Autonomous Systems, allows the robot to learn to swim in a manner similar to that of beavers on its own.
“We offer a biologically inspired reinforcement learning control method to describe the mobility of underwater robots in this study,” Gang Chen, one of the paper’s authors, told TechXplore. “This method is mostly based on one of our prior studies of beaver motion, which was published in Springer Link’s Journal of Intelligent and Robotic Systems.”
Underwater robots, like the one developed by Chen and his colleagues, are nonlinear systems with complicated hydrodynamics. Modeling their movements accurately can be a difficult and time-consuming operation that requires large computing resources.
In contrast to previous methods for guiding the mobility of underwater robots, Chen and his colleagues’ solution does not require the incorporation of sophisticated motion models based on hydrodynamics. This is due to the fact that it uses simplified joint angle representations to dynamically simulate beaver swimming action. These joint representations make training the model easier while also decreasing the robot’s inefficient motions.
“Our strategy executes the robot’s swimming control as swiftly and operably as possible by combining reinforcement learning with the mechanics driving beaver swimming behavior,” Chen added. “It’s most noticeable and unique benefit is that it can avoid constructing sophisticated motion control models and quickly accomplish swimming control of a robot.”
Chen and his colleagues used a single-legged robotic platform to test their beaver-inspired reinforcement learning-based strategy in a series of trials. Their approach produced effective beaver-like swimming motions that improved the robot’s locomotion, and their results were highly promising.
The technology developed by this group of researchers could be utilized to improve the performance and movements of other one-legged robots that operate in water in the future. Furthermore, their study may serve as a model for developing similar ways to controlling the movements of other underwater robots.
“We intend to improve the construction and performance of the beaver-like swimming robot in the future,” Chen added. “We’d also like to look into ways to use reinforcement learning to improve the intelligence behind robotic swimming motions, focusing not just on the robot’s swimming velocity but also on swimming stability, trajectory planning, and obstacle avoidance, all in a real underwater environment.”
