Technology, Robotic, Artificial IntelligenceJuly 26, 2023Researchers develop efficient machine-learning technique for enhanced robot control in dynamic environmentsResearchers from MIT and Stanford University have developed a machine-learning technique that allows for more efficient and effective control of robots in dynamic environments. The researchers' approach incorporates certain structure from control theory into the process of learning a model, resulting in the creation of stabilizing controllers that can better handle complex dynamics caused by factors like wind impacts on flying vehicles. By jointly learning the system's dynamics and control-oriented structures from data, the approach naturally creates controllers that function more effectively in real-world scenarios. Unlike traditional methods, which require separate learning of a controller, this technique immediately extracts an effective controller from the learned model. Additionally, the approach requires fewer data points, enabling faster learning and adaptation to rapidly changing conditions. The researchers believe that their method can be applied to various types of robots, from drones to autonomous vehicles, and could lead to the development of more physically interpretable models for improved performance.
Source: Massachusetts Institute of Technology