Differences

This shows you the differences between two versions of the page.

--- teaching:mfe:ia [2011/03/23 16:07]
mdorigo [Self-organized task allocation in swarm robotics]
+++ teaching:mfe:ia [2011/03/24 10:40]
mdorigo
@@ Line 182: / Line 182: @@
 This project is about single player games (puzzles) and the design of algorithms for tackling hard combinatorial optimisation problems.
-Example puzzles are: <a href="http://en.wikipedia.org/wiki/Light_Up">Light Up</a>, <a href="http://en.wikipedia.org/wiki/Mastermind_(board_game)">Mastermind</a>, <a href="http://en.wikipedia.org/wiki/Minesweeper_(video_game)">Minesweeper</a>, etc.
+Example puzzles are: [[http://en.wikipedia.org/wiki/Light_Up|Light Up]], [[http://en.wikipedia.org/wiki/Mastermind_(board_game)|Mastermind]], [[http://en.wikipedia.org/wiki/Minesweeper_(video_game)|Minesweeper]], etc.
 The student will learn how to design and implement a Stochastic Local Search algorithm to solve NP-complete puzzles. The student will also learn how to analyse the performaces of the algorithm and perform statistically sound comparisons with the other algorithms available in literature.
@@ Line 280: / Line 280: @@
   * Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Marco Dorigo, Eliseo Ferrante, Ali Emre Turgut (IRIDIA)
-===== A comparison of decision-making strategies for adaptive foraging in swarm robotics =====
-Group of social insects are able to efficiently find the (shortest) path to the a food source and even to differentiate between the quality of two food sources. Studies with ants showed that this mechanism is driven by the perception of stimuli from chemical substances like pheromone. Moreover ants are able to collectively modify their choices if there are changes in the environment, that is, if a source becomes better than another. These ideas have been a source of inspiration for several algorithms in swarm robotics which solves a similar problem (retrieval of objects) by using different types of stimuli such as the encounter rate of objects.
-The goal of this project is to perform a study on how to solve a foraging task in which robots have to choose between staying at the nest or go foraging for different energy sources. The optimal strategy might change over time. What happens if all the robots go to the best source? Will these "traffic jams" slow the process? Is it possible to avoid this problem? What if source quality changes over time? The study will be conducted only in simulation and will concern comparing different approaches and different metrics to measure stimuli.
-Required skills: The candidates should be acquainted with C++ programming and have a working knowledge of the English language.
-  * Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Marco Dorigo, Eliseo Ferrante, Manuele Brambilla (IRIDIA)
 ===== Kaleidoscope: Creating temporal motion patterns in a swarm of robots =====
@@ Line 300: / Line 291: @@
+===== Flocking and avoidance of concave obstacles =====
+In swarm robotics, taking inspiration from biological
+systems such as flock of birds, a number of strategies for coordinated
+navigation of large swarms have been proposed. By coordinated
+navigation we mean that each robot in the swarm is able to locally avoid
+collisions with the neighbors, while at the global level the swarm
+behaves as a single organism. In cluttered environments, in which many
+obstacles are present, it is still an open problem how to maintain the
+cohesion of the swarm. In addition, when obstacles are concave and
+large enough to house a large portion of the swarm, the latter may get stuck
+in the concavity, preventing navigation from continuing. This project
+aims to study a set of effective strategies to tackle this
+problem. The student will run experiments with simulated and real
+robots. The most important required skill is a good knowledge of C and
+C++. The working language is English.
+* Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Marco Dorigo, Carlo Pinciroli, Eliseo Ferrante (IRIDIA)
+===== Automatic fitness function definition in evolutionary robotics =====
+Evolutionary robotics is a fascinating approach to the design of robot controllers that takes inspiration from natural evolution.
+In order to obtain a robot that is able to perform a desired task, the evolutionary robotics approach considers a population of robots that evolves in time. Each robot is characterized by a genotype that defines somehow its behavior. Each robot is evaluated according to a fitness function that measures the ability of the robot to perform the desired task. Robots with a low fitness are eliminated. Robots with a high fitness remain in the population and generate offsprings -- e.g., robots with a similar genotype obtained via mutation and/or cross-over. Through this process, generation by generation, the evolutionary robotics approach is able to obtain robots that present higher and higher fitness and that are therefore able to perform the desired task more and more effectively.
+One of the main open problems in evolutionary robotics is that the definition of an appropriate fitness function is a very complex, labor-intensive, and time-consuming activity that requires the attention of an expert researcher.
+The goal of this master thesis is to devise an automatic method to define a fitness function in order to obtain a robot that is able to perform a desired task. This automatic method will be based on machine learning and metaheuristic algorithms. In particular, it will draw ideas from the fields of reinforcement learning and of on-line adaptation of parameters in optimization algorithms.
+Required skills: The candidates should be acquainted with C/C++ programming and have a working knowledge of the English language.
+* Contact: [[http://iridia.ulb.ac.be/~mbiro|Mauro Birattari]], Marco Dorigo, Vito Trianni (IRIDIA)
+===== Evolution of Modular Controllers for Simulated and Real Robots =====
+The goal of this master thesis is investigating how modularity in a robot controller can influence the quality of the behaviours obtained through artificial evolution.
+Similarly to the nervous system that can be divided in central and peripheral, the project will study a modular architecture for neural network controllers. The peripheral modules encode the information coming from the sensory subsytems or going to the motor apparatus. The central system encodes the behavioural rules that map sensations to actions. The project will study methods to develop the peripheral modules by maximising the information transfer from the sensory input and to the motor output, on the basis of measures derived from Information Theory.
+The project will involve experimental activities with both simulated and real robots, and will investigate both individual and collective behaviours.
+Required skills: The candidates should be acquainted with C/C++ programming and have a working knowledge of the English language.
+* Contact: [[http://iridia.ulb.ac.be/~vtrianni|Vito Trianni]], Marco Dorigo (IRIDIA)