DOCSLIB.ORG

Challenges and Opportunities of Evolutionary Robotics

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Challenges and Opportunities of Evolutionary Robotics In Proceedings of the Second International Conference on Computational Intelligence, Robotics and Autonomous Systems (CIRAS), Singapore, December 2003. Challenges and Opportunities of Evolutionary Robotics D. A. Sofge, M. A. Potter, M. D. Bugajska, A. C. Schultz Navy Center for Applied Research in Artificial Intelligence Naval Research Laboratory Washington, D.C. 20375, USA {sofge, mpotter, magda, schultz}@aic.nrl.navy.mil Abstract The need for ER arises from the fact that as robotic sys- tems and the environments into which they are placed Robotic hardware designs are becoming more complex as increase in complexity, the difficulty of programming the variety and number of on-board sensors increase and their control systems to respond appropriately increases to as greater computational power is provided in ever- the point where it becomes impracticable to foresee every smaller packages on-board robots. These advances in possible state of the robot in its environment. Evolution- hardware, however, do not automatically translate into ary algorithms are used to generate control algorithms better software for controlling complex robots. Evolu- using the Darwinian principle of survival-of-the-fittest. A tionary techniques hold the potential to solve many diffi- population of controllers is maintained and evolved by cult problems in robotics which defy simple conventional evaluating individual control systems based on a measure approaches, but present many challenges as well. Nu- of how well they achieve desired characteristics such as merous disciplines including artificial life, cognitive sci- executing appropriate behaviors at appropriate times. ence and neural networks, rule-based systems, behavior- Only the fitter members of the population survive and based control, genetic algorithms and other forms of evo- pass the characteristics that made them successful on to lutionary computation have contributed to shaping the future generations. The ultimate goal is to produce the current state of evolutionary robotics. This paper pro- best possible controller given some design criteria. As vides an overview of developments in the emerging field of discussed in the following sections, this approach has evolutionary robotics, and discusses some of the opportu- proven very successful in a wide variety of challenging nities and challenges which currently face practitioners in robotics domains. the field. The remainder of this paper will provide a sampling of recent development efforts in evolutionary robotics re- 1. Introduction search with the goal of showing both the opportunities presented by the use of evolutionary techniques for solv- The field of evolutionary robotics has emerged in recent ing difficult problems in robotics, but also in showing years as the application of artificial evolution to the de- some of the challenges of applying these techniques. velopment of robotic systems. While most of the work in evolutionary robotics has focused on the development of control systems for autonomous mobile robots, some re- 2. Evolved Controllers for Autonomous searchers have used the techniques to evolve robotic Mobile Robots hardware configurations and even robot body parts along with the controllers. A key objective in evolutionary robotics is to evolve be- havior-based controllers for autonomous mobile robots Evolutionary robotics (ER) has its origins in several dis- [8,9]. Autonomous mobile robots often incorporate both ciplines including artificial life [1], cognitive science and reactive and longer-term planning components in order to neural networks [2,3,4], behavior-based control [5], ge- accommodate goal-driven behaviors. The reactive por- netic programming [6] and genetic algorithms [7]. Cur- tion of the controller may be encoded in a variety of rent practitioners of ER incorporate techniques from a forms. Common choices include stimulus-response rules, variety of disciplines to achieve the desired result, often a neural networks, and state-machines. The planning stage robotic control system for an autonomous mobile robot may be represented as a series of goal states. Behavior- (or other autonomous system) which exhibits a set of de- based controllers are thus driven by a combination of cur- sired behaviors, and for which those behaviors were ac- rent state, as determined by current sensor readings and quired “automatically” (i.e. without custom programming possibly short-term memory, and goal(s). The controller of each individual behavior). attempts to match its current state readings and goal, which together comprise the stimulus part, and then to produce the appropriate output, or response. 2.1. Evolved Rule-Based Control techniques [14]. Evolutionary algorithms have proven quite successful in training recurrent neural network- Work by Grefenstette and Schultz [10] resulted in the based controllers for autonomous mobile robots [9]. application of the SAMUEL learning system to evolve stimulus-response rules to produce a reactive control sys- Potter et al. [15] demonstrated the evolution of neural tem for autonomous mobile robots. Behaviors achieved network controllers for multiple robots engaged in shep- using this system include tracking, navigation, and obsta- herding a sheep robot. The goal of this work was to ex- cle avoidance. SAMUEL maintains a population of can- amine the effects of evolving a single homogeneous con- didate behaviors which are evaluated in a simulated ro- troller for a group of autonomous mobile robots perform- botic environment. The population is scored, mutation ing a collective herding task, versus coevolving separate and crossover operators applied, and the population is heterogeneous controllers, and to determine if the com- adjusted to remove the lesser scoring rule sets. The proc- plexity of the task favored homogeneous or heterogene- ess is run for a number of generations until a stopping ous control. The heterogeneous controllers were produced criteria is met, at which point the best evolved controller using a cooperative coevolutionary architecture [16] in is uploaded to the robot hardware for validation and test- which the controller for each robot is evolved in a sepa- ing. This technique proved successful for evolving reac- rate genetically-isolated species. It was found that het- tive controllers for autonomous mobile robots. Schultz et erogeneous controllers are indeed advantageous when the al. [12] demonstrated the evolution of rule-based control- task can be decomposed into subtasks that can be solved lers for learning complex robotic behaviors by evolving by robots specializing in substantially different skill sets. the behavior for a shepherd robot to coerce a sheep robot Otherwise homogeneous controllers have an advantage into a corral. This technique and was extended to evolv- due to their generality. ing controllers for simulated autonomous aircraft and autonomous underwater vehicles [7,11]. Quinn et al. [17] described another experiment in which neural network controllers were evolved for a team of real Challenges with evolving rule-based controllers include robots. The objective was to study the capability of the determining how to map continuous inputs and outputs to robots to learn formation forming behaviors starting from discrete state variables, establishing appropriate interme- random positions and using a minimal sensor set consist- diate and goal states, determining how many production ing of 4 IR sensors on each robot. rules are required, and performing conflict resolution. 3. Hyper-Redundant Robot Control 2.2. Evolved Neural Network Based Control One of the greatest challenges in robotics is the design of An alternative representation used by many researchers in control systems for robots with high degrees of freedom, evolutionary robotics is artificial neural networks, which particularly if many of the degrees of freedom are cou- have a number of characteristics that are desirable from pled. A key example of this is a highly segmented ser- an evolutionary robotics perspective. Neural networks are pentine or snake-like robotic arm (Fig. 1). This is an ex- relatively insensitive to noise in the environment since the ample of a hyper-redundant robot, where there are many output of each node is typically a function of the input possible kinematic solutions to achieve the same end- from a variety of sources. This characteristic also pro- effector position or trajectory. duces a smooth search space with a well-behaved map- ping between changes to the network and changes in the resultant network behavior. Neural networks also natu- rally handle continuous input and can produce either con- tinuous or discrete output as desired [9]. A key advantage of using evolutionary algorithms for producing robotic neural network based controllers is that the evolutionary techniques may be used equally well Figure 1. Hyper-redundant robotic manipulator with feed-forward or recurrent networks [13]. Recurrent networks offer many advantages for dynamic control sys- Sofge [18] used evolutionary algorithms to generate the tems because they allow recent state information to be inverse kinematics for the hyper-redundant robotic ma- combined with current state information in the decision- nipulator. The continuous work space of the robot was making process. In effect, the recurrent connections pro- discretized into a grid of regularly spaced points. A popu- vide a kind of short-term memory capability for the neural lation of genomes was created such that each genome network so that decisions
Load more

Recommended publications
  • Evolving Symbolic Controllers
    Evolving Symbolic Controllers Nicolas Godzik1, Marc Schoenauer1, and Mich`ele Sebag2 1 Projet Fractales, INRIA Rocquencourt, France 2 LRI, Universit´eParis-Sud, France Published in G. Raidl et al., eds, Applications of Evolutionary Computing, pp 638-650, LNCS 2611, Springer Verlag, 2003. Abstract. The idea of symbolic controllers tries to bridge the gap be- tween the top-down manual design of the controller architecture, as advo- cated in Brooks’ subsumption architecture, and the bottom-up designer- free approach that is now standard within the Evolutionary Robotics community. The designer provides a set of elementary behavior, and evolution is given the goal of assembling them to solve complex tasks. Two experiments are presented, demonstrating the efficiency and show- ing the recursiveness of this approach. In particular, the sensitivity with respect to the proposed elementary behaviors, and the robustness w.r.t. generalization of the resulting controllers are studied in detail. 1 Introduction There are two main trends in autonomous robotics. There are two main trends in autonomous robotics. The first one, advocated by R. Brooks [2], is a human- specified deterministic approach: the tasks of the robot are manually decom- posed into a hierarchy of independent sub-tasks, resulting in the the so-called subsumption architecture. On the other hand, evolutionary robotics (see e.g. [13]), is generally viewed as a pure black-box approach: some controllers, mapping the sensors to the actua- tors, are optimized using the Darwinian paradigm of Evolutionary Computation; the programmer only designs the fitness function. However, the scaling issue remains critical for both approaches, though for arXiv:0705.1244v1 [cs.AI] 9 May 2007 different reasons.
    [Show full text]
  • A Model for Virtual Reconfigurable Modular Robots Thomas Breton, Yves Duthen
    A Model for Virtual Reconfigurable Modular Robots Thomas Breton, Yves Duthen To cite this version: Thomas Breton, Yves Duthen. A Model for Virtual Reconfigurable Modular Robots. 2011. hal- 01298411 HAL Id: hal-01298411 https://hal.archives-ouvertes.fr/hal-01298411 Preprint submitted on 5 Apr 2016 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. A Model for Virtual Reconfigurable Modular Robots Thomas Breton1 and Yves Duthen1 1VORTEX Research Team, IRIT UMR 5505 University of Toulouse, France [email protected], [email protected] Abstract and communication methods. We propose here a model to accomplish different tasks such as motion planning, This paper presents a model for virtual reconfigurable modu- object displacement, and structure reconfiguration; evolving lar robots in order to evolve artificial creatures, able of self- adaptation to the environment as well as good adjustment to a controller (a neural network) by the mean of a genetic various given tasks. For this purpose, a simulator has been algorithm. entirely developed with the assistance of a physics engine to represent force activities. One of the most crucial points It is true that any consideration taken in modular robotic in modular robot construction is the choice of module type, complexity and diversity.
    [Show full text]
  • Evolutionary Robotics in Two Decades: a Review
    Sadhan¯ a¯ Vol. 40, Part 4, June 2015, pp. 1169–1184. c Indian Academy of Sciences Evolutionary robotics in two decades: A review SAMEER GUPTA and EKTA SINGLA∗ School of Mechanical, Materials and Energy Engineering, IIT Ropar, Rupnagar 140001, India e-mail: [email protected]; [email protected] MS received 29 August 2014; revised 1 January 2015; accepted 11 February 2015 Abstract. Evolutionary robotics (ER) has emerged as a fast growing field in the last two decades and has earned the attention of a number of researchers. Principles of biological evolution are applied in the form of evolutionary techniques for solv- ing the complicated problems in the areas of robotic design and control. The diversity and the intensity of this growing field is presented in this paper through the contribu- tions made by several researchers in the categories of robot controller design, robot body design, co-evolution of body and brain and in transforming the evolved robots in physical reality. The paper discusses some of the recent achievements in each of these fields along with some expected applications which are likely to motivate the future research. For the quick reference of the readers, a digest of all the works is presented in the paper, spanning the years and the areas of the research contributions. Keywords. Evolutionary robotics; evolutionary control; robot morphology; body– brain design. 1. Introduction With the advent of bio-inspired computational techniques (Brooks 1986; Beer & Gallagher 1992; Haykin 1994; Ram et al 1994), the interests of many roboticists bend towards the utility of these techniques in the complicated designs and control of robots.
    [Show full text]
  • Evolving Robot Empathy Through the Generation of Artificial Pain in An
    Evolving Robot Empathy through the Generation of Artificial Pain in an Adaptive Self-Awareness Framework for Human-Robot Collaborative Tasks Muh Anshar Faculty of Engineering and Information Technology University of Technology Sydney This dissertation is submitted for the degree of Doctor of Philosophy March 2017 Bismillahirrahmanirrahim All Praise and Gratitude to the Almighty God, Allah SWT, for His Mercy and Guidance which have given me strength and tremendous support to maintain my motivation from the very beginning of my life journey and into the far future. I would like to dedicate this thesis to my love ones, my wife and my son, Nor Faizah & Abdurrahman Khalid Hafidz for always being beside me which has been a great and undeniable support throughout my study. CERTIFICATE OF ORIGINAL AUTHORSHIP This thesis is the result of a research candidature conducted jointly with another University as part of a collaborative Doctoral degree. I certify that the work in this thesis has not previously been submitted for a degree nor has it been submitted as part of requirements for a degree except as part of the collaborative doctoral degree and/or fully acknowledged within the text. I also certify that the thesis has been written by me. Any help that I have received in my research work and the preparation of the thesis itself has been acknowledged. In addition, I certify that all information sources and literature used are indicated in the thesis. Signature of Student: Date: 13 March 2017 Muh Anshar March 2017 Acknowledgements I would like to acknowledge and thank my Principal Supervisor, Professor Mary-Anne Williams for her great dedication, support and supervision throughout my PhD journey.
    [Show full text]
  • Swarm Robotics Distributed Embodied Evolutionary Robotics
    Swarm Robotics Distributed Embodied Evolutionary Robotics (Evolutionary) Swarm Robotics: a gentle introduction Inaki˜ Fernandez´ Perez´ [email protected] www.loria.fr/˜fernandi ISAL Student Group ECAL 2017 September 8th 2017 Universite´ de Lorraine, LARSEN Team, Inria Nancy, France 1 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics So::: what is a robot swarm? Large/huge set of Focus in collective Real robots are cool::: simple agents dynamics but simulation works too! 2 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics So::: what is a robot swarm? Large/huge set of Focus in collective Real robots are cool::: simple agents dynamics but simulation works too! 2 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics Where to start::: many approaches By hand: ∼ engineering approach [Brambilla et al., 2013] Clones: classical EA with copies on each robot (homogeneous) [Tuci et al., 2008] Coevolution: either cooperative or competitive (heterogeneous) [Gomes et al., 2016] (distributed) Embodied Evolution runs onboard on each robot (heterogeneous) [Watson et al., 2002, Fernandez´ Perez´ et al., 2017] 3 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics What goals? ALife models to understand biology/test biological hypothesis ALife tools to build systems that can solve a problem for us 4 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics What robotic tasks? Navigation Flocking Item collection Foraging Shepherding ::: 5 / 10 Swarm Robotics Distributed Embodied Evolutionary Robotics
    [Show full text]
  • Simulating the Evolution of Soft and Rigid-Body Robots
    Simulating the Evolution of So and Rigid-Body Robots Sam Kriegman Collin Cappelle∗ Francesco Corucci MEC Labx MEC Lab e BioRobotics Institute University of Vermont University of Vermont Scuola Superiore Sant’Anna Burlington, VT, USA Burlington, VT, USA Pisa, Italy [email protected] [email protected] Anton Bernatskiy Nick Cheney Josh C. Bongard MEC Lab Creative Machines Lab MEC Lab University of Vermont Cornell University University of Vermont Burlington, VT, USA Ithaca, NY, USA Burlington, VT, USA ABSTRACT here two high-level Python wrappers around two dierent simula- In evolutionary robotics, evolutionary methods are used to optimize tors: Open Dynamics Engine (ODE) and Voxelyze. Our goal in this robots to dierent tasks. Because using physical robots is costly paper is to convey from experience what kinds of things are di- in terms of both time and money, simulated robots are generally cult/easy to instantiate in physics engines in general and how our used instead. Most physics engines are wrien in C++ which can user-friendly modules at least partially alleviate this, and how they be a barrier for new programmers. In this paper we present two may be extended in the future through open-source collaborations. Python wrappers, Pyrosim and Evosoro, around two well used simulators, Open Dynamics Engine (ODE) and Voxelyze/VoxCAD, 2 RIGID BODY ROBOTS which respectively handle rigid and so bodied simulation. Python Rigid body dynamics engines are generally what is most thought is an easier language to understand so more time can be spent of when one thinks of a simulator. ey are 3D engines where on developing the actual experiment instead of programming the every body in the simulation is ‘rigid’, meaning the body cannot simulator.
    [Show full text]
  • Evolutionary Robotics
    Evolutionary Robotics IAR Lecture 13 Barbara Webb Basic process Population of genomes, e.g. Decode each into robot binary strings, tree structures controller and/or morphology, e.g. weights in neural net, position of Produce new set of sensors genomes, e.g. breed, crossover, mutate Place in environment and run Use fitness to select for Evaluate behaviour reproduction, e.g. only if using a fitness function achieved task, or best e.g. achieve task, speed, individuals, or proportional time survived, find mate to fitness score Motivation • Lack of design methods that will ensure the right dynamics emerge from the environment-robot-task interaction • Automate the trial-and-error approach • Avoid preconceptions in design • Allow self-organising processes to discover novel and efficient solutions • Good enough for biology (and might help us understand biology) ‘Typical’ example Floreano & Mondada (1996): evolving Braitenberg-type control for a Khepera robot to move around maze • Eight IR sensor input units, feed-forward to two motor output units with recurrent connections • Standard sigmoidal ANN n 1 y f w x , where f (x) i ij j kx j 1 e • Genome – bit string encoding weight values • Fitness function: V(1 v)(1i) where i is highest IR value, V vleft vright v vleft vright • Population of 80, each tested for approx 30s • Copied proportional to fitness, then random paired single point crossover and mutation (prob.=0.2) • 100 generations, get smooth travel round maze Similar approach has been used to evolve controllers for more complex
    [Show full text]
  • Evolutionary Developmental Soft Robotics Towards Adaptive and Intelligent Machines Following Nature’S Approach to Design
    Evolutionary Developmental Soft Robotics Towards adaptive and intelligent machines following Nature’s approach to design Francesco Corucci, PhD November 16th, 2017 - ShanghAI Lectures Motivations: diversity, complexity, sophistication F. Corucci Evolutionary Developmental Soft Robotics 2 Motivations: intelligent and adaptive behavior Camouflage Creativity Skills Reasoning, cognition F. Corucci Evolutionary Developmental Soft Robotics 3 Motivations Can we automatically design a wealth of artificial systems that are as sophisticated, adaptive, robust, intelligent, for a wide variety of tasks and environments? F. Corucci Evolutionary Developmental Soft Robotics 4 Adaptivity, robustness, intelligence State of the art robots still lack many of these features Keep failing outside controlled environments (where they are most needed) DARPA Robotics Challenge Finals, 2015 F. Corucci Evolutionary Developmental Soft Robotics 5 Biologically inspired robotics (biorobotics) Cheetah robot, MIT Bat robot, Brown Soft fish, MIT OCTOPUS, SSSA RoboBees, Harvard Lampetra, SSSA Plantoid robot, IIT ECCE robot F. Corucci Evolutionary Developmental Soft Robotics 6 Biologically inspired robotics: Soft Robotics F. Corucci Evolutionary Developmental Soft Robotics 7 Biologically inspired robotics: pros and cons Pros: New technologies and design principles New knowledge related to the biological model (sometimes) Insights related to the intelligence of particular species (sometimes) Cons: Requires a lot of human knowledge and careful engineering Focuses on very specific
    [Show full text]
  • DART: Diversity-Enhanced Autonomy in Robot Teams
    DART: Diversity-enhanced Autonomy in Robot Teams Nora Ayanian Abstract This paper defines the research area of Diversity-enhanced Autonomy in Robot Teams (DART), a novel paradigm for the creation and design of policies for multi-robot coordination. While current approaches to multi-robot coordination have been successful in structured, well understood environments, they have not been successful in unstructured, uncertain environments, such as disaster response. The reason for this is not due to limitations in robot hardware, which has advanced significantly in the past decade, but in how multi-robot problems are solved. Even with significant advances in the field of multi-robot systems, the same problem- solving paradigm has remained: assumptions are made to simplify the problem, and a solution is optimized for those assumptions and deployed to the entire team. This results in brittle solutions that prove incapable if the original assumptions are inval- idated. This paper introduces a new multi-robot problem-solving paradigm which relies on a diverse set of control policies that work together synergistically to make multi-robot systems more resilient in unstructured and uncertain environments. 1 Introduction The field of multi-robot systems (MRS) is growing at a rapid pace. Research in MRS spans many different areas, including automated delivery [1–3], surveillance [4], and disaster response [5, 6]. There have also been many successful demonstrations of increasing numbers of robots [7–9, 11–13]. MRS have also been successfully deployed in the field including in warehousing [15], manufacturing [16], and enter- tainment [17]. While these outcomes show the promise of MRS, the environments in which MRS have been successful are highly controlled, and some are highly in- strumented, enabling precise tuning of controllers and nearly perfect knowledge of environmental conditions.
    [Show full text]
  • Bootstrapping Q-Learning for Robotics from Neuro-Evolution Results Matthieu Zimmer, Stephane Doncieux
    Bootstrapping Q-Learning for Robotics from Neuro-Evolution Results Matthieu Zimmer, Stephane Doncieux To cite this version: Matthieu Zimmer, Stephane Doncieux. Bootstrapping Q-Learning for Robotics from Neuro-Evolution Results. IEEE Transactions on Cognitive and Developmental Systems, Institute of Electrical and Electronics Engineers, Inc, 2017, 10.1109/TCDS.2016.2628817. hal-01494744 HAL Id: hal-01494744 https://hal.archives-ouvertes.fr/hal-01494744 Submitted on 23 Mar 2017 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS, VOL. X, NO. Y, JANUARY 2016 1 Bootstrapping Q-Learning for Robotics from Neuro-Evolution Results Matthieu Zimmer, Stephane Doncieux, Member, IEEE Abstract—Reinforcement learning problems are hard to solve require exploring large and continuous search spaces and in a robotics context as classical algorithms rely on discrete typically rely on demonstrations by an expert to bootstrap, representations of actions and states, but in robotics both are or on non-random initial policies, or else on dedicated policy continuous. A discrete set of actions and states can be defined, but it requires an expertise that may not be available, in particular representations [2,3,4].
    [Show full text]
  • Evolutionary Robotics
    Introduction to Evolutionary Robotics Robotics Class, 2017, Prof. Laschi Teaching Assistant: Francesco Corucci f.corucci santannapisa.it http://sssa.bioroboticsinstitute.it/user/1507 @f_corucci Who am I • 4th year PhD Student in BioRobotics at The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa (IT) (Advisor: Prof. Laschi) • Visiting researcher, Morphology, Evolution and Cognition Lab, Vermont Complex Systems Center, University of Vermont (USA) (Advisor: Prof. Josh Bongard) TA for this course, 4h on Evolutionary Robotics + projects on this topic More info: http://sssa.bioroboticsinstitute.it/user/1507 F. Corucci - Introduction to Evolutionary Robotics Part of this material is inspired by or adapted from: • Floreano, Matiussi, “Bio-Inspired Artificial Intelligence - Theories, Methods, and Technologies”, MIT press (2008). • Pfeifer, Bongard, “How the body shapes the way we think: a new view of intelligence”. MIT press (2006). • “Evolutionary Robotics” class (CS206) at University of Vermont (USA), Prof. J. Bongard • LudoBots, a Reddit-based MOOC on Evolutionary Robotics https://www.reddit.com/r/ludobots/wiki/index http://www.reddit.com/r/ludobots http://www.meclab.org/ F. Corucci - Introduction to Evolutionary Robotics Outline • Introduction and motivation • Evolutionary Algorithms • Human-competitive design and “perverse instantiation” • Developmental encodings and co-evolution of artificial brains and bodies • The “reality gap” problem Appendix: • VoxCad, multi-material physics engine for soft-robot analysis, design and evolution F. Corucci - Introduction to Evolutionary Robotics Introduction and motivation F. Corucci - Introduction to Evolutionary Robotics Biorobotics: sometimes successful… Passive dynamic walker, Mc Geer 1990 F. Corucci - Introduction to Evolutionary Robotics Biorobotics: sometimes successful… Koh et al. 2015, Science F. Corucci - Introduction to Evolutionary Robotics …sometimes not so much Robots failing at the Darpa Robotic Challenge 2015 (IEEE Spectrum) F.
    [Show full text]
  • The AAAI 1999 Mobile Robot Competitions and Exhibitions
    AI Magazine Volume 21 Number 3 (2000) (© AAAI) Articles The AAAI 1999 Mobile Robot Competitions and Exhibition Lisa Meeden, Alan Schultz, Tucker Balch, Rahul Bhargava, Karen Zita Haigh, Marc Böhlen, Cathryne Stein, and David Miller I The Eighth Annual Mobile Robot Competition place was captured by University of Arkansas’s and Exhibition was held as part of the Sixteenth ELECTRO, with Douglas Blank as the adviser and National Conference on Artificial Intelligence in students Jared Hudson, Brian Mashburn, and Orlando, Florida, 18 to 22 July. The goals of these Eric Roberts. robot events are to foster the sharing of research In Hors d’oeuvres, Anyone? the entrants and technology, allow research groups to showcase served finger-food items at the large reception their achievements, encourage students to enter robotics and AI fields at both the undergraduate for conference attendees at the American Asso- and graduate level, and increase awareness of the ciation for Artificial Intelligence festival. The field. The 1999 events included two robot contests; objective was to unambiguously demonstrate a new, long-term robot challenge; an exhibition; interaction with the spectators. The teams’ and a National Botball Championship for high robots were encouraged to display distinctive school teams sponsored by the KISS Institute. Each personalities; allow two-way communication of these events is described in detail in this article. with the spectators; and, if necessary, physical- The contests require entrants to tackle a set of ly nudge spectators out of the way to better established tasks in a competitive environ- serve a wide area of the room. The Hors d’oeu- ment.
    [Show full text]