14th International Conference on Automatic Control and Soft Computing

The use of renewable energy, such as solar energy, experienced a great impulse during the second half of the seventies just after the first big oil crisis. At that time economic issues were the most important factors and the interest in these types of processes decreased when the oil prices fell. There is a renewed interest in the use of renewable energies nowadays driven by the need of reducing the high environmental impact produced by the use of fossil energy systems. There are two main drawbacks of energy systems: a) the resulting energy costs are not yet competitive and b) solar energy is not always available when needed. Considerable research efforts are being devoted to techniques which may help to overcome these drawbacks, control is one of those techniques. A thermal solar power plant basically consists of a system where the solar energy is collected, then concentrated and finally transferred to a fluid. The thermal energy of the hot fluid is then used for different purposes such as generating electricity, the desalination of sea water etc. While in other power generating processes, the main source of energy (the fuel) can be manipulated as it is used as the main control variable, in solar energy systems, the main source of power which is solar radiation cannot be manipulated and furthermore it changes in a seasonal and on a daily base acting as a disturbance when considering it from a control point of view. Solar plants have all the characteristics needed for using advanced control strategies able to cope with changing dynamics, (nonlinearities and uncertainties). As fixed PID controllers cannot cope with some of the mentioned problems, they have to be detuned with low gain, producing sluggish responses or if they are tightly tuned they may produce high oscillations when the dynamics of the process vary, due to environmental and/or operating conditions changes. The use of more efficient control strategies resulting in better responses would increase the number of operational hours of the plants. The talk describes the main solar thermal plants, the control problems involved and how control systems can help in increasing their efficiency. Some illustrative examples are given.

The project aims at designing and constructing a twin wing, tailless flapping wing robot of the size of a hummingbird with emphasis on biomimetic flight. The project started in 2010; the various steps of the project include: wing aerodynamics, flapping wing mechanism, stability and control (the system is naturally unstable), flight dynamics, power consumption, flight simulation and flight testing. The vehicle weights a little more than 20 grams for a wing span of 21 cm; the flapping frequency is 22 Hz. The maiden flight took place in June 2016. Since then, the research has been focussed on improving the flight quality, exploring the aeroelastic coupling and extending the flight time. A short video can be seen here

In this workshop, we'll employ Deep Reinforcement Learning to train a biped robot through simulation to walk safely and optimally following a straight line.

Designing autonomous systems (robots, vehicles, virtual assistants...) requires solving complex optimal control problems that are difficult to undertake because it's hard to define a control strategy or the objectives for each variable.

Machine Learning makes it possible to train "black box" algorithms with example data to tackle sophisticated tasks. Lying at the intersection with Game Theory, Reinforcement Learning is probably the branch with the most promising future in Automatic Control. By means of it, "agents" learn "policies" (control strategies) through trial and error. In order for these policies and the training process to be sophisticated enough, it's often useful to implement them with deep neural networks.

In our example, we'll start with a 3D physical model of the robot in Simscape Multibody™. We'll craft an accurate enough model of the environment and its rewards in Simulink. With Deep Learning Toolbox™, we'll design neural networks to codify a Reinforcement Learning algorithm and the agent's policy. Then we'll run several simulations in a parallel cluster and, once the agent is trained, we'll see how to deploy it to a programmable controller in the real robot through automatic C++ code generation.

Registration details:

1. Until June 30, you must make the registration on this website;
2. The masterclass will be presented in real-time using the "Microsoft Teams" platform;
3. A certificate of participation will be issued for all the persons enrolled in the masterclass as long as the registration in the webpage is made up to the above deadline;
4. A temporary MATLAB license will be provided for all the attendees.