Memory Modelling for Artificial Companions
The aim of this project is to investigate computational principles for
learning and memory as well as their application to robot companions. In
order to evaluate these principles in practise, an appropriate system is
developed and tested in suitable scenarios. In order to store
information efficiently and in a compact way, the research focuses on
highly important memory processes such as association, generalisation
and aspects of forgetting. In particular, methods for associative
learning and incremental clustering are applied to autonomously learn
from unlabelled data. By combining these learning techniques, the
information captured by various sensors can be generalised and related
to each other. Moreover, data from actuators can be incorporated, which
allows a robot to react to different input stimuli. The developed system
strictly avoids the inclusion of any hard-coded or domain knowledge and
aims for an easy applicability.
Swarm Robotics - Intelligent control and evolutionary strategies applied to multirobot systems
The main objective of this project is the development and the evaluation of bio-inspired models, including swarm behavior and evolutionary computation combined with machine learning techniques to provide autonomy and efficiency in the coordination of robot groups performing task in critical activities. The proposal algorithms must support high degrees of scalability, flexibility and fault tolerance. The robot group control must occur in autonomy and flexible way, independent of any kind of human supervision. There are many fields where a single agent is not sufficient or enough to fulfill a task. Tasks like cleaning nuclear residuals, cleaning chemical accidents, forest fire combat or even constructions, agriculture, hostile environment exploration, security and critical missions might be better accomplished using a group of agents. The use of robotic agents instead of human beings, in these applications, may add security, reliability and efficiency. These characteristics are going to be evaluated both in a virtual environment, using physical simulation tools as with real robots.
Synchronisation, networks of coupled phase oscillators and behaviour: applying neurodynamics to evolutionary robotics
This project focuses on the interplay between the collective interaction of neuronal oscillators and behaviour. We make use an extended version of a widely known neural network model of phase interacting oscillators, the Kuramoto model to different evolutionary robotics tasks commonly studied for attesting minimally cognitive behaviours. We believe that exploring the simulated brain/body/environment system could contribute to unveil important mechanisms of the neural system which may not be easily identifiable in living organisms. Moreover, the research outcomes could inspire the design of new robotic controllers as well as shed light into many different research areas, from the comprehension of the role of oscillatory properties in some diseases to the establishment of new parallel computing architectures.
The Use of Unmanned Aerial Vehicles Integrated with Wireless Sensor Network for Agricultural Application
This project is conducted in collaboration with key researchers from the University of São Paulo (Brazil) and the University of Bern (Switzerland). Dr. Patricia A. Vargas describes an architecture based on unmanned aerial vehicles (UAVs) which can be employed to implement a control loop for agricultural applications where UAVs are responsible for spraying chemicals on crops. The process of applying the chemicals deployed on the crop field is controlled by means of feedback obtained from the wireless sensor networks (WSN).