Papers by Fernando Fernandez
Abstract: Before multi-robot teams will ever become widely used in practice, we believe that adva... more Abstract: Before multi-robot teams will ever become widely used in practice, we believe that advances must be made in the development of mechanisms that enable the robot teams to autonomously generate and adapt cooperative behaviors. With the current state of the art, the implementation of cooperative behaviors on physical robot teams requires expert behavior programming and experimentation, followed by extensive
Catheterization and Cardiovascular Diagnosis, 1991
A percutaneous mitral balloon valvotomy (PMBV) was attempted on 190 patients with fluoroscopic gu... more A percutaneous mitral balloon valvotomy (PMBV) was attempted on 190 patients with fluoroscopic guidance of atrial septal puncture for transseptal catheterization; in 3 cases, the procedure could not be performed. The left atrium was always reached on the first attempt, when the relationship of the Brockenbrough needle to the aortic catheter was previously observed in 3 fluoroscopic views: anteroposterior, 45" right anterior oblique, and lateral. The atrial septal puncture site was located immediately below the aortic valve level, probably in the fossa ovalis, for the first 80 patients, and at mid distance between the aortic valve level and the diaphragm for the last 110. Hemodynamic data were similar in both groups.
Planning, execution and monitoring of physical rehabilitation therapies with a robotic architecture
Studies in health technology and informatics, 2015
Traditional methods of rehabilitation require continuous attention of therapists during the thera... more Traditional methods of rehabilitation require continuous attention of therapists during the therapy sessions. This is a hard and expensive task in terms of time and effort. In many cases, the therapeutic objectives cannot be achieved due to the overwork or the difficulty for therapists to plan accurate sessions according to the medical criteria. For this purpose, a wide range of studies is opened in order to research new ways of rehabilitation, as in the field of social robotics. This work presents the current state of the THERAPIST project [1]. Our main goal is to develop a cognitive architecture which provides a robot with enough autonomy to carry out an upper-limb rehabilitation therapy for patients with physical impairments, such as Cerebral Palsy and Obstetric Brachial Plexus Palsy.
Lecture Notes in Computer Science, 2003
The design of nearest neighbour classifiers is very dependent from some crucial parameters involv... more The design of nearest neighbour classifiers is very dependent from some crucial parameters involved in learning, like the number of prototypes to use, the initial localization of these prototypes, and a smoothing parameter. These parameters have to be found by a trial and error process or by some automatic methods. In this work, an evolutionary approach based on Nearest Neighbour Classifier (ENNC), is described. Main property of this algorithm is that it does not require any of the above mentioned parameters. The algorithm is based on the evolution of a set of prototypes that can execute several operators in order to increase their quality in a local sense, and emerging a high classification accuracy for the whole classifier.
Lecture Notes in Computer Science, 2000
Reinforcement learning has proven to be a set of successful techniques for finding optimal polici... more Reinforcement learning has proven to be a set of successful techniques for finding optimal policies on uncertain and/or dynamic domains, such as the RoboCup. One of the problems on using such techniques appears with large state and action spaces, as it is the case of input information coming from the Robosoccer simulator. In this paper, we describe a new mechanism for solving the states generalization problem in reinforcement learning algorithms. This clustering mechanism is based on the vector quantization technique for signal analog-to-digital conversion and compression, and on the Generalized Lloyd Algorithm for the design of vector quantizers. Furthermore, we present the VQQL model, that integrates Q-Learning as reinforcement learning technique and vector quantization as state generalization technique. We show some results on applying this model to learning the interception task skill for Robosoccer agents.
Dalton transactions (Cambridge, England : 2003), Jan 21, 2007
A new family of functionalized ligands derived from norborn-5-ene-2,3-dicarboxylic anhydride has ... more A new family of functionalized ligands derived from norborn-5-ene-2,3-dicarboxylic anhydride has been used in Suzuki C-C cross-couplings between aryl boronic acids and aryl bromide derivatives in [BMI][PF(6)] (BMI=1-n-butyl-3-methyl-imidazolium), using palladium acetate as catalytic precursor. High conversions and yields are obtained when amine ligands containing hydroxy groups are involved. TEM analyses after catalysis show the formation of small nanoparticles, in contrast to the agglomerates observed when nanoparticles are intentionally preformed, with a consequent decrease in the catalytic activity in the latter case. Some tests, including the correlation effect between solvent and ligand, are carried out to try to identify the true nature of the catalyst. All the results obtained suggest that formation of nanoparticles is required to lead to a catalytically active system.
One of the most studied areas of human reasoning from a computational point of view has been plan... more One of the most studied areas of human reasoning from a computational point of view has been planning. Classical Automated Planning (AP) deals with the generation of ordered sets of actions that, when correctly executed, allow an agent (or set of agents) to transit from an initial state to a final state where a set of goals are satisfied. We are interested on building intelligent systems that sense, reason (plan) and execute plans in the real world. Since most real environments are stochastic and/or dynamic, we focus on planners that handle the extra complexity of considering non-determinism and/or dynamism. Currently, the two major approaches to deal with these environments consist on explicit reasoning on probabilistic domain models, or, in the other extreme, reasoning only with full deterministic models (removing all probabilistic information), execution of generated plans, and replanning when needed. In this paper, we propose a variation of the second approach. Our technique, Variable Resolution Planning (VRP), performs a detailed analysis of the near future (first actions in the plan). After a given planning horizon, it abstracts away details as it reasons about the far future. The advantages of this approach are that it: is faster than regular fullfledged planning (both in the probabilistic or deterministic settings); does not spend much time on the far future possibilities that will not be reached anyway, since in most cases it will need to replan before finding the end of the plan; takes into account some main components of the far future (as an improvement over pure reactive systems); and, from a cognitive perspective, plausibly is closer to how humans plan in these kinds of environments.
Planning a sequence of actions is particularly difficult in stochastic environments, where action... more Planning a sequence of actions is particularly difficult in stochastic environments, where actions execution might fail, which in turn prevents the execution of the rest of the plan. Also, in many domains, agents cannot wait a long time for plan generation. In this paper, we propose the use of Variable Resolution Planning (VRP). The main idea is based on the fact that if the domain is stochastic it is usually not worth computing a valid (sound) long plan, since most of it will not really be used. So, VRP generates a plan where the first k actions are applicable if the environment does not change, while the rest of the plan might not be necessarily applicable, since it has been generated using an abstraction by removing some predicates. Also, planning with abstraction requires less computation time than computing regular applicable plans. Experimental results show the advantages of this approach in several domains over a standard planning approach.
IBACOP and IBACOP2 planner
Learning predictive models to configure planning portfolios
Complex robotic tasks require the coordination of a considerable amount of skills. This is genera... more Complex robotic tasks require the coordination of a considerable amount of skills. This is generally achieved generating and executing action plans that fulfill the preconditions of the given objective. These tasks can be highly dynamic, since the appearance of new objects or unexpected situations is a constant during the plan execution. In this context, robot control systems require the capability of managing a suitable world model (creating, removing or retyping dynamically objects as a result of the plan execution), and the capability of monitoring and replanning when unexpected situations are detected. In this paper we introduce a general-purpose architecture for autonomous mobile robots providing these features. The architecture allows to generate planning applications since it integrates planning, re-planning, monitoring and learning capabilities, and, at the same time, manages a consistent graph-like world model. Finally, we present some preliminary results of the deployment of such architecture in an advertisement promoting robot domain.
A rehabilitation therapy usually derives from general goals set by the medical expert, who reques... more A rehabilitation therapy usually derives from general goals set by the medical expert, who requests the patient to attend sessions during a certain time period in order to help him regaining mobility, strength and/or flexibility. The therapist must transform these general goals manually into a set of exercises distributed over different rehabilitation sessions that compose the complete therapy plan, taking into account the patient clinical conditions and a predetermined session and therapy time. This becomes a hard task and might lead to rigid schedules which not always accomplish the desired achievement level of therapeutic objectives established by the physician and could have a negative impact on the patients' engagement in the therapy. In this paper we present a method based on Automated Planning for the automatic generation of therapy plans for patients suffering obstetric braquial plexus palsy, in response to a given set of therapy goals. The ultimate purpose is to project the therapy plans to robots that can help patients achieving a better performance, showing them how to do exercises properly.
Lecture Notes in Computer Science, 2004
One of the most important issues in educational systems is to define effective teaching policies ... more One of the most important issues in educational systems is to define effective teaching policies according to the students learning characteristics. This paper proposes to use the Reinforcement Learning (RL) model in order for the system to learn automatically sequence of contents to be shown to the student, based only in interactions with other students, like human tutors do. An initial clustering of the students according to their learning characteristics is proposed in order the system adapts better to each student. Experiments show convergence to optimal teaching tactics for different clusters of simulated students, concluding that the convergence is faster when the system tactics have been previously initialised.
Materials Letters, 2001
Textured strontium ferrite thin films has been grown at room temperature using a Nd:YAG laser. Th... more Textured strontium ferrite thin films has been grown at room temperature using a Nd:YAG laser. The spectroscopic study of the produced plasma revealed that the expansion velocities of the species are of the order of 106 cm/s, which could explain the obtained texture. The stoichiometric analysis shows a small oxygen reduction in the films due to the absence of a
MARL-Ped: A multi-agent reinforcement learning based framework to simulate pedestrian groups
Simulation Modelling Practice and Theory, 2014
Lecture Notes in Computer Science, 2012
In this paper, the calibration of a framework based in Multiagent Reinforcement Learning (RL) for... more In this paper, the calibration of a framework based in Multiagent Reinforcement Learning (RL) for generating motion simulations of pedestrian groups is presented. The framework sets a group of autonomous embodied agents that learn to control individually its instant velocity vector in scenarios with collisions and friction forces. The result of the process is a different learned motion controller for each agent. The calibration of both, the physical properties involved in the motion of our embodied agents and the corresponding dynamics, is an important issue for a realistic simulation. The physics engine used has been calibrated with values taken from real pedestrian dynamics. Two experiments have been carried out for testing this approach. The results of the experiments are compared with databases of real pedestrians in similar scenarios. As a comparison tool, the diagram of speed versus density, known as fundamental diagram in the literature, is used.
Autonomous Agents and Multi-Agent Systems, 2014
In this paper, a new Multi-agent Reinforcement Learning (MARL) approach is introduced for the sim... more In this paper, a new Multi-agent Reinforcement Learning (MARL) approach is introduced for the simulation of pedestrian groups. Unlike other solutions, where the behaviors of the pedestrians are coded in the system, in our approach the agents learn by interacting with the environment. The embodied agents must learn to control their velocity, avoiding obstacles and the other pedestrians, to reach a goal inside the scenario. The main contribution of this paper is to propose this new methodology that uses different iterative learning strategies, combining a Vector Quantization (state space generalization) with the Q-learning algorithm (VQQL). Two algorithmic schemas, Iterative VQQL (ITVQQL) and Incremental (INVQQL), which differ in the way of addressing the problems, have been designed and used with and without transfer of knowledge. These algorithms are tested and compared with the VQQL algorithm as a baseline in two scenarios where agents need to solve well-known problems in pedestrian modeling. In the first, agents in a closed room need to reach the unique exit producing and solving a bottleneck. In in the second, two groups of agents inside a corridor need to reach their goal that is placed in opposite sides (they need to solve the crossing). In the first scenario, we focus on scalability, use metrics from the pedestrian modeling field, and compare with the Helbing's social force model. The emergence of collective behaviors, that is, the shell-shaped clogging in front of the exit in the first scenario, and the lane formation as a solution to the problem of the crossing, have been obtained and analyzed. The results demonstrate that the proposed schemas find policies that carry out the tasks, suggesting that they are applicable and generalizable to the simulation of pedestrians groups.
Emergent Collective Behaviors in a Multi-agent Reinforcement Learning Pedestrian Simulation: A Case Study
Lecture Notes in Computer Science, 2015
2012 IEEE Conference on Evolving and Adaptive Intelligent Systems, 2012
Robot motor control learning is currently one of the most active research areas in robotics. Many... more Robot motor control learning is currently one of the most active research areas in robotics. Many learning techniques have been developed for relatively simple problems. However, very few of them have direct applicability in complex robotic systems without assuming prior knowledge about the task, mainly due to three facts. Firstly, they scale badly to continues and high dimensional problems. Secondly, they need too many real robot-environment interactions. Finally, they are not capable of adapting to environment or robot dynamic changes. In order to overcome these problems, we have developed a new algorithm capable of finding from scratch open-loop state-action trajectory solutions by mixing sample-based tree kinodynamic planning with dynamic model learning. Some results demonstrating the viability of this new type of approach in the cart-pole swing-up task problem are presented.
Lecture Notes in Computer Science, 2003
The paper shows t he archi tecture of t he RL ATES syst em, an Adaptive and Intelligent Education... more The paper shows t he archi tecture of t he RL ATES syst em, an Adaptive and Intelligent Educational System th at u ses th e Re inforcement Le arning model (RL) in order to learn to teach each student individually, being adapted to their learning needs in each moment of the interaction. This papers is focused on the i nterface modul e of RL ATES, descr ibing how t he st udent coul d navi gate through the system interface and how this interface adjusts the page contents according to the user l earning needs. F or t his adapt ation, t he syst em changes t he links appearance of the page and the presentation of the system knowledge.
Uploads
Papers by Fernando Fernandez