In this project, the first release of BT Studio has been developed. BT Studio is an open-source tool crafted for the development of robotic applications. Its primary objective is to facilitate the quick deployment of behavior tree-based robotic applications within ROS 2. In BT Studio, a robotic app is defined as an graphical tree coupled with actions scripted in Python, which the tool then translates into a ROS 2 package. This process circumvents the unnecessary complexities often associated with ROS-specific configurations, offering developers a more streamlined approach. The generated robotics application may be run locally or even from the browser inside a docker container with one button click.

The automotive world does not escape the advances of robotics and AI. one of the great dreams in the history of humanity Since cars circulate in our world, autonomous cars are becoming increasingly popular every day closer to being a reality. In some countries, autonomous cars are already going circulate through large cities, even doing taxi services. The world of autonomous driving is also related to the racing motorsport, with the existence of some competitions in which one one of the most important elements, if not the most, of a racing car, which is the pilot is not a person but a computer capable of understanding and analyzing the car environment to drive himself as fast as possible around track.

This TFG focuses on the development of autonomous driving behaviors based on reinforcement learning in racing circuits, in order to obtain reactive and appropriate behaviors in different scenarios using a robot with the necessary characteristics to carry out these behaviors, the AWS Deepracer.
The AWS Deepracer league is a competition where participants try to build an autonomous behavior in a small racing car using artificial intelligence. In this competition, participants are provided with a small robot with autonomous capabilities, the AWS Deepracer promoted by Amazon, together with a simulated environment in which the participants must program the robot with one goal: be the fastast one!



Check the video below to see more details about the system:

Artificial intelligence is already helping in the education field to understand the needs of each student and improve the learning methodologies for a better education. Neural networks such as PoseNet or FaceMesh can help to understand the dynamic of the class without expose the student's privacy. In collaboration with Tknika a deep learning vision-based framework has been developed to detect and analyze emotions, behaviours, movements and talk-action of the students in a regular class.



This work focuses on developing a system for detecting and analyzing emotions and behaviors based on AI on a teaching environment. Main goal of the project is to provide useful information to teachers about student performance. This system detection is based on algorithms that have been designed to perform a visual analysis of images and videos using pose and face neural networks for emotion detection. After all the analysis the information obtained is adapted to use as a guide and help for teachers, who will be able to understand visually the different events that occurred during a class in relation to a certain students. Privacy has been treated as a central pillar in this project, that's why no images are saved, neither audio of the environment is analyzed.

Autonomous done navigation is getting attention in different fields during last years. Applications focused on people-rescue, surveillance, infrastructure inspection are being demanded nowadays, but drone must be operated autonomously by itself.
The focus of the study is on Unmanned Air Vehicles (UAV), whose use covers from rescues in inaccessible areas, to inspections in the industry, or inventoried, among many other examples.In parallel, we have the study of the Radio Frequency (RF) spectrum, which covers a large part of the modern world, such as the use of communications mobile phones, or in broadcasting using frequency bands such as FM radio. Thus, Understanding and working on signal propagation can introduce us to new useful applications for the world.

To evaluate properly this project, a FRISS model has been implemented in python and create a ROS node to export this feature to the ROS ecosystem. Check the PDF above for more information.

The objective of this work is to developed a solution based on reinfocemente learning (Q-learning) that is able to learn how to reach a RF transmmiter at outdoor/indoor environment. Static and dynamic environments have been evaluated with obstácules in a realistic way.

Check the video below to see more details about the system:

In this project an end-to-end vision-based deep learning approach for an autonomous car to drive in traffic is presented. In this context the car should both keep the lane and also keep a safety distance with the front vehicle, if any. The imitation learning approach has been followed, creating a supervised dataset with the autopilot of CARLA simulator in different traffic conditions which include front vehicles. A variant of the PilotNet model has been developed, with additional dropout and the previous speed as input, and it has been (re)trained with such in-traffic dataset. The experimental results show that the model successfully drives the car in traffic conditions without losing performance in free road conditions. It also keeps safety distance with the front cars and even properly generalizes to several types of front vehicles.

A plugin for measuring the distance to the front vehicle has been created and added to Behavior Metrics, an open-source autonomous driving assessment tool.

Robotics and Artificial Intelligence (AI) are areas in continuous expansion that covers many different fields. One of the sectors that has benefited the most of the development of this industry is medicine. People who present any injury, disability, who need help to walk or who have suffered an accident cerebrovascular disease, require assistance to accelerate their recovery and improve their quality of life. Exoskeletons play a fundamental role by allowing them to perform rehabilitation exercises quickly and efficiently, relieving the physical load of the therapists involved. Due to this, designing and developing new programs of rehabilitation that are integrated into these devices is absolutely necessary, with the purpose of addressing the different problems that patients face in your recovery process.



In collaboration with the company Marsi Bionics, a new exercise has been designed rehabilitation for an exoskeleton. After a study of pre-existing information on the device, the necessary content was extracted and a new control system based on the movement of the squats. Thanks to this system, patients will be able to perform exercises of this type, which is important considering note that it is not a common movement in current devices. Besides, am evaluation has been made with pose neural network models, in order to provide additional assistance in situations where the exoskeleton is not available.

Finally, in this project it has been possible to develop a solid and stable solution of the proposed exercise, as well as a comparison of two neural network models verifying that, indeed, although the results are more accurate using the exoskeleton, could be used for support and additional help.