Scientists from ITMO have developed a way to automate the design of robots using the MCTS (Monte Carlo Tree Search) algorithm. This algorithm, which once beat grandmasters at chess, is now being used to generate designs for robotic devices. As a result, creating a new project takes only a couple of hours, while the manual process can take a week. The scientific results were presented at the IEEE Intelligent Robots and Systems (IROS) conference.
Designing a robot requires taking into account many factors, such as shape, internal structure, kinematics and dynamics, as well as materials and sensors. Engineers often encounter conflicts between these parameters, which complicates the task. As a result, they spend a lot of time searching for the optimal balance and sorting through thousands of possible options. This makes the development process long and dependent on the personal experience and intuition of the designer.
To simplify the task, ITMO graduate students adapted the MCTS algorithm, which goes through all possible combinations and checks their effectiveness in the simulator. The engineer sets the initial parameters, such as the number of fingers or the search step, after which the algorithm creates and tests the models. Each is rated based on performance. As one of the authors of the project, graduate student Kirill Zharkov, noted, the algorithm is capable of processing thousands of combinations and selecting the best solutions, which makes it more effective than traditional methods. Already tested designs of gripping devices created by the algorithm have successfully coped with the task of grasping objects. However, the developers plan to refine the code to reduce the gap between virtual experiments and real conditions.
