Researchers create a machine-learning system that can learn to manage a robot effectively, resulting in higher performance with less data.

It might be used to more successfully and efficiently drive a robot, such as a drone or autonomous vehicle, in dynamic environments where conditions can change quickly. This technique might enable a drone to closely follow a downhill skier despite being buffeted by high winds, a robotic free-flyer to haul various items in space, and an automated vehicle to learn to correct for slick road conditions to avoid entering a skid.

Learning process

The researchers' method incorporates a specific structure from control theory into the learning process of a model in a way that efficiently regulates complicated dynamics, including those brought on by wind impacts on the trajectory of a flying object. This structure is a clue that can help direct system control to use one way of thinking about it.

This structure allows researchers to immediately extract an effective controller from the model, unlike other machine learning methods that require controller development or learning. With this framework, their algorithm can learn an efficient controller with less information. It could speed up how their learning-based control system improves performance in dynamic settings.

Learning to use a controller

Even when researchers know how to describe everything about the system, determining the optimum approach to direct a robot to a specific task can be a complex problem. A controller, for example, is the logic that allows a drone to follow a particular trajectory. This controller would instruct the drone on changing its rotor forces to compensate for the effect of winds, which could cause it to deviate from a steady path to its destination.

This drone is a dynamic system, which means it is a physical system that changes over time. As it flies through the surroundings, its position and velocity fluctuate. Engineers can create a controller by hand if the system is simple enough. Hand-modelling a system captures a specific structure based on the physics of the system. For example, if a robot were manually modelled using differential equations, the relationship between velocity, acceleration, and force would be captured. The rate of change in velocity over time is dictated by the mass of the robot and the forces applied to it.

Aerodynamic effects

However, the system could be more complicated to be precisely modelled by hand. Instead, researchers would collect data on the drone's position, velocity, and rotor speeds over time and apply machine learning to fit a model of this dynamical system to the data. However, these techniques only sometimes learn a control-based structure. This structure helps us know how to configure the rotor speeds to best steer the drone's motion over time. Many existing approaches employ data to learn a separate controller for the system after modelling the dynamical system.

Conclusion

The researchers also discovered that their strategy was data-efficient, which indicates that it performed well even with limited data. It could effectively represent a highly dynamic rotor-driven vehicle with only 100 data points. With smaller datasets, methods that used numerous learned components saw their performance degrade significantly faster. Their technique could be precious when a drone or robot needs to know quickly in rapidly changing conditions due to its efficiency.

Furthermore, their method is broad and might be applied to various dynamical systems, from robotic arms to free-flying spacecraft in low-gravity situations.

Image source: Unsplash