Doncieux S., Bredeche N., Mouret J-B. (eds.) New Horizons in Evolutionary Robotics

Extended Contributions from the 2009 EvoDeRob Workshop. — Springer-Verlag Berlin Heidelberg, 2011. XVI, 225 p. — ISBN: 978-3-642-18271-6, e-ISBN: 978-3-642-18272-3, DOI 10.1007/978-3-642-18272-3 — (Studies in Computational Intelligence,Volume 341).Evolutionary Algorithms (EAs) now provide mature optimization tools that have successfully been applied to many problems, from designing antennas to complete robots, and provided many human-competitive results.In robotics, the integration of EAs within the engineer’s toolbox made tremendous progress in the last 20 years and proposes new methods to address challenging problems in various setups: modular robotics, swarm robotics, robotics with non-conventional mechanics (e.g. high redundancy, dynamic motion, multi-modality), etc.This book takes its roots in the workshop on "New Horizons in Evolutionary Design of Robots" that brought together researchers from Computer Science and Robotics during the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS-2009) in Saint Louis (USA). This book features extended contributions from the workshop, thus providing various examples of current problems and applications, with a special emphasis on the link between Computer Science and Robotics. It also provides a comprehensive and up-to-date introduction to Evolutionary Robotics after 20 years of maturation as well as thoughts and considerations from several major actors in the field.This book offers a comprehensive introduction to the current trends and challenges in Evolutionary Robotics for the next decade.Evolutionary Robotics: Exploring New Horizons
St´ephane Doncieux, Jean-Baptiste Mouret, Nicolas Bredeche, Vincent PadoisA Brief Introduction to Evolutionary Computation
Whento Use ER Methods?
Absence of Optimal Method
Knowledge of Fitness Function Primitives
Knowledge of Phenotype Primitives
Where and How to Use EA in the Robot Design Process?
Mature Techniques: ParameterTuning
Current Trend: Evolutionary Aided Design
Current Trend: Online Evolutionary Adaptation
Long Term Research: Automatic Synthesis
Frontiers of ER and Perspectives
Reality Gap
Fitness Landscape and Exploration
Genericity of Evolved Solutions
A Roboticist Point of View
Discussion
Good Robotic Engineering Practices
Good Experimental Sciences PracticesInvited Position PapersThe ‘What’, ‘How’ and the ‘Why’ of Evolutionary Robotics
Josh Bongard
The What of Embodiment
The How of Embodiment
The Why of Embodiment
Why Consider Topological Change to a Robot’s Body Plan?
Why Evolve Robot Body Plans Initially at a Low Resolution?
Why Allow Body Plans to Change during Behavior Optimization?Why Evolutionary Robotics Will Matter
Kenneth O. Stanley
Joining the Mainstream
Bridging the Gap
Realizing the PromiseEvolutionary Algorithms in the Design of Complex Robotic Systems
Philippe BidaudParticularities of the RoboticSystemDesign
Parameters and Evaluation of Robotic Systems
Evolutionary Algorithms in the Robotic System Design
Kinematic Design of Robot Manipulators
Modular Locomotion System Design
Inverse Model Synthesis
Multi-objective Task Based Design of Redundant Systems
Flexible Building Block Design of Compliant MechanismsRegular ContributionsEvolving Monolithic Robot Controllers through Incremental Shaping
Joshua E. Auerbach, Josh C. BongardLearning Multiple Behaviors with a Monolithic Controller
Specialization in a Morphologically Homogeneous RobotEvolutionary Algorithms to Analyze and Design a Controller for a Flapping Wings Aircraft
St´ephane Doncieux, Mohamed HamdaouiMethod
Experimental Setup
Results
Discussion and FutureWorkOn Applying Neuroevolutionary Methods to Complex Robotic Tasks
Yohannes Kassahun, Jose de Gea, Jakob Schwendner,
Frank KirchnerCase Study 1: Augmented Neural Network with Kalman Filter (ANKF)
The αβ Filter
Evolving ANKF
Comparison of Number of Parameters to be Optimized for ANKF and Recurrent Neural
Networks
Results Obtained for ANKF on the Double Pole Balancing without Velocities Benchmark
Case Study 2: Incremental Modification of Fitness Function
Quadrocopter
Control Architecture Developed for the Quadrocopter Using the Principles of Behavior
Based Systems
Incremental Modification of Fitness Function
Experimentsand Results
Task Decomposition with a Definition of a Single Global Fitness Function Is Not Necessarily
Sufficient for Solving Complex Robot TasksEvolutionary Design of a Robotic Manipulator for a Highly Constrained Environment
S. Rubrecht, E. Singla, V. Padois, P. Bidaud, M. de BroissiaCase Study
Genetic Algorithm and Implementation
Genetic Algorithm
Genome
TrajectoryTracking
Control Law
Indicators
Results
Design with Simple Trajectory
Design with Complex Trajectory
Conclusions and Future WorksFutureWorksA Multi-cellular Based Self-organizing Approach for Distributed Multi-Robot Systems
Yan Meng, Hongliang Guo, Yaochu JinBiologicalBackground
The Approach
The GRN-Based Dynamics
Convergence Analysis of System Dynamics
The Evolutionary Algorithm for Parameter Tuning
SimulationandResults
Case Study 1: Multi-robots Forming a Unit Circle
Case Study 2: Multi-robots Forming a Unit Square
Case Study 3: Self-reorganization
Case Study 4: Robustness Tests to Sensory Noise
Case Study 5: Self-adaptation to Environmental Changes
Conclusion and FutureWorksNovelty-Based Multiobjectivization
Jean-Baptiste MouretRelated Work
Novelty Search
Multi-Objective Evolutionary Algorithms
Multiobjectivization
Method
Experiment
Fitness Function and Distance between Behaviors
Variants
Expected Results
Experimental Parameters
Results
Average Fitness
Convergence Rate
Exploration
Conclusion and DiscussionEmbedded Evolutionary Robotics: The (1+1)-Restart-Online Adaptation Algorithm
Jean-Marc Montanier, Nicolas BredecheExtendingthe(1+1)-OnlineEA
Limitsof (1+1)-Online
The(1+1)-Restart-Online Algorithm
Experiments and Results
Hardware Set-Up
Experimental Set-Up
Experimental Results
Hall-of-Fame Analysis
Real Robot Experiment
Conclusion and PerspectivesAutomated Planning Logic Synthesis for Autonomous Unmanned Vehicles in Competitive Environments with Deceptive Adversaries
Petr Svec, Satyandra K. GuptaUSV System Architecture
USV Virtual Sensor Models
Planning Architecture
Planning Logic Synthesis
Test Mission
Synthesis Scheme
Planning Logic Components Evolution
Computational Experiments
General Setup
ResultsMajor Feedback Loops Supporting Artificial Evolution in Multi-modular Robotics
Thomas Schmickl, J¨urgen Stradner, Heiko Hamann, Lutz Winkler, Karl CrailsheimArtificial Homeostatic Hormone System
Artificial Genome
Feedback1: Classic Control
Feedback2: Learning
Feedback3: Evolution
Feedback4: Controller Morphogenesis
Feedback 5: Robot Organism Morphogenesis
Feedback6: BodyMotion
Step 1: The First Oscillator
Step 2: Motion of Bigger Organisms
Step 3: Motion of More Complex Organisms
DiscussionEvolutionary Design and Assembly Planning for Stochastic Modular Robots
Michael T. Tolley, Jonathan D. Hiller, Hod LipsonTarget Structure Evolution
Stochastic Fluidic Assembly System Model
Assembly Algorithm