Simulation and control of distributed robot search teams

This article describes the simulation of distributed autonomous robots for search and rescue operations. The simulation system is utilized to perform experiments with various control strategies for the robot team and team organizations, evaluating the comparative performance of the strategies and organizations. The objective of the robot team is to, once deployed in an environment (floor-plan) with multiple rooms, cover as many rooms as possible. The simulated robots are capable of navigation through the environment, and can communicate using simple messages. The simulator maintains the world, provides each robot with sensory information, and carries out the actions of the robots. The simulator keeps track of the rooms visited by robots and the elapsed time, in order to evaluate the performance of the robot teams. The robot teams are composed of homogenous robots, i.e., identical control strategies are used to generate the behavior of each robot in the team. The ability to deploy autonomous robots, as opposed to humans, in hazardous search and rescue missions could provide immeasurable benefits.     

Evolving non-trivial behaviors on real robots: A garbage collecting robot

Recently, a new approach involving a form of simulated evolution has been proposed to build autonomous robots. However, it is still not clear if this approach is adequate for real life problems. In this paper we show how control systems that perform a non-trivial sequence of behaviors can be obtained with this methodology by “canalizing” the evolutionary process in the right direction. In the experiment described in the paper, a mobile robot was successfully trained to keep clear an arena surrounded by walls by locating, recognizing, and grasping “garbage” objects and by taking collected objects outside the arena. The controller of the robot was evolved in simulation and then downloaded and tested on the real robot. We also show that while a given amount of supervision may canalize the evolutionary process in the right direction the addition of unnecessary constraints can delay the evolution of the desired behavior.     

Evolutionary control of bipedal locomotion in a high degree-of-freedom humanoid robot: first steps

This article describes a methodology, together with an associated series of experiments employing this methodology, for the evolution of walking behavior in a simulated humanoid robot with up to 20 degrees of freedom. The robots evolved in this study learn to walk smoothly in an upright or near-upright position and demonstrate a variety of different locomotive behaviors, including “skating,” “limping,” and walking in a manner curiously reminiscent of a mildly or heavily intoxicated person. A previous study demonstrated the possible potential utility of this approach while evolving controllers based on simulated humanoid robots with a restricted range of movements. Although walking behaviors were developed, these were slow and relied on the robot walking in an excessively stooped position, similar to the gait of an infirm elderly person. This article extends the previous work to a robot with many degrees of freedom, up to 20 in total (arms, elbows, legs, hips, knees, etc.), and demonstrates the automatic evolution of fully upright bipedal locomotion in a humanoid robot using an accurate physics simulator. This work was presented in part at the 11th International Symposium on Artificial Life and Robotics, Oita, Japan, January 23–25, 2006     

GP-Sumo: Using genetic programming to evolve sumobots

We describe the evolution—via genetic programming—of control systems for real-world, sumo-fighting robots—sumobots, in adherence with the Robothon rules: Two robots face each other within a circular arena, the objective of each being to push the other outside the arena boundaries. Our robots are minimally equipped with sensors and actuators, the intent being to seek out good fighters with this restricted platform, in a limited amount of time. We describe four sets of experiments—of gradually increasing difficulty—which also test a number of evolutionary methods: single-population vs. coevolution, static fitness vs. dynamic fitness, and real vs. dummy opponents.     

Genetic learning of virtual team member preferences

Virtual team members do not have complete understanding of other team members’ preferences, which makes team coordination somewhat difficult and time consuming. Traditional approaches for team coordination require a lot of inter-agent electronic communication and often result in wasted effort. Methods that reduce inter-agent communication and conflicts are likely to increase productivity of virtual teams. In this research, we propose an evolutionary genetic algorithm (GA) based intelligent agent that learns a team member preferences from past actions, and develops a team-coordination schedule by minimizing schedule conflicts between different members serving on a virtual team. Using a discrete event simulation methodology, we test the proposed intelligent agent on different virtual teams of sizes two, four, six and eight members. The results of our experiments indicate that the GA-based intelligent agent learns individual team member preferences and generates a team-coordination schedule at a lower inter-agent communication cost.     

Evolutionary learning of a fuzzy controller for wall-following behavior in mobile robotics

The design of fuzzy controllers for the implementation of behaviors in mobile robotics is a complex and highly time-consuming task. The use of machine learning techniques such as evolutionary algorithms or artificial neural networks for the learning of these controllers allows to automate the design process. In this paper, the automated design of a fuzzy controller using genetic algorithms for the implementation of the wall-following behavior in a mobile robot is described. The algorithm is based on the iterative rule learning approach, and is characterized by three main points. First, learning has no restrictions neither in the number of membership functions, nor in their values. In the second place, the training set is composed of a set of examples uniformly distributed along the universe of discourse of the variables. This warrantees that the quality of the learned behavior does not depend on the environment, and also that the robot will be capable to face different situations. Finally, the trade off between the number of rules and the quality/accuracy of the controller can be adjusted selecting the value of a parameter. Once the knowledge base has been learned, a process for its reduction and tuning is applied, increasing the cooperation between rules and reducing its number.     

A review of procedures to evolve quantum algorithms

There exist quantum algorithms that are more efficient than their classical counterparts; such algorithms were invented by Shor in 1994 and then Grover in 1996. A lack of invention since Grover’s algorithm has been commonly attributed to the non-intuitive nature of quantum algorithms to the classically trained person. Thus, the idea of using computers to automatically generate quantum algorithms based on an evolutionary model emerged. A limitation of this approach is that quantum computers do not yet exist and quantum simulation on a classical machine has an exponential order overhead. Nevertheless, early research into evolving quantum algorithms has shown promise. This paper provides an introduction into quantum and evolutionary algorithms for the computer scientist not familiar with these fields. The exciting field of using evolutionary algorithms to evolve quantum algorithms is then reviewed.

Motion coordination for industrial robotic systems with redundant degrees of freedom

This work investigates how to distribute in an optimum fashion the desired movement of the end-effector of an industrial robot with respect to the workpiece, when there are redundant degrees of freedom, such as a positioning table. The desired motion is given as a series of acceleration functions in respective time intervals. The constraints of the optimisation are the available acceleration limit of axes, such as the table axes, the upper bounds to velocity and displacement of each axis and the avoidance of singular point areas of the robot, as defined by its manufacturer. The optimisation criterion is minimum total work for the motion. A genetic algorithm was used to solve the problem. The fitness function of the genetic algorithm calls a kinematics and dynamics simulation model of the robotic installation constructed in Matlab™, in order to compute the work consumed and to check possible violation of constraints. Examples of straight line and circular movement are given to prove the concept. Results are encouraging, yet demand on computing power is high.     

Scaling of program functionality

The distribution of fitness values (landscapes) of programs tends to a limit as the programs get bigger. We use Markov chain convergence theorems to give general upper bounds on the length of programs needed for convergence. How big programs need to be to approach the limit depends on the type of the computer they run on. We give bounds (exponential in N, N log N and smaller) for five computer models: any, average or amorphous or random, cyclic, bit flip and four functions (AND, NAND, OR and NOR). Programs can be treated as lookup tables which map between their inputs and their outputs. Using this we prove similar convergence results for the distribution of functions implemented by linear computer programs. We show most functions are constants and the remainder are mostly parsimonious. The effect of ad-hoc rules on genetic programming (GP) are described and new heuristics are proposed. We give bounds on how long programs need to be before the distribution of their functionality is close to its limiting distribution, both in general and for average computers. The computational importance of destroying information is discussed with respect to reversible and quantum computers. Mutation randomizes a genetic algorithm population in generations. Results for average computers and a model like genetic programming are confirmed experimentally.     

Multiobjective topology optimization of structures using genetic algorithms with chromosome repairing

In this work, a genetic algorithm (GA) for multiobjective topology optimization of linear elastic structures is developed. Its purpose is to evolve an evenly distributed group of solutions to determine the optimum Pareto set for a given problem. The GA determines a set of solutions to be sorted by its domination properties and a filter is defined to retain the Pareto solutions. As an equality constraint on volume has to be enforced, all chromosomes used in the genetic GA must generate individuals with the same volume value; in the coding adopted, this means that they must preserve the same number of “ones” and, implicitly, the same number of “zeros” along the evolutionary process. It is thus necessary: (1) to define chromosomes satisfying this propriety and (2) to create corresponding crossover and mutation operators which preserve volume. Optimal solutions of each of the single-objective problems are introduced in the initial population to reduce computational effort and a repairing mechanism is developed to increase the number of admissible structures in the populations. Also, as the work of the external loads can be calculated independently for each individual, parallel processing was used in its evaluation. Numerical applications involving two and three objective functions in 2D and two objective functions in 3D are employed as tests for the computational model developed. Moreover, results obtained with and without chromosome repairing are compared.     

遗传算法及其进化硬件设计研究

遗传算法(Genetic Algorithms，GAs)是进化计算中的重要领域．也是人工智能迅速发展的重要领域，它是一类模拟自然进化过程和达尔文“适者生存”的算法。这一算法企图通过使用诸如交叉、变异等算子从先前的候选解中生成最佳解。进化硬件(Evolvable Hardware，EHW)则是受自然进化得到灵感、设计硬件系统的新型设计方法，它是进化计算技术和电子硬件结合的产物。本文给出GAs在计算机模型中的基本概念．讨论进化再配置硬件的设计，最后给出几个进化实验的例子。     

An information-theoretic landscape analysis of neuro-controlled embodied organisms

Recently, there has been a lot of interest in evolving controllers for both physically simulated creatures as well as for real physical robots. However, a range of different ANN architectures are used for controller evolution, and, in the majority of the work conducted, the choice of the architecture used is made arbitrarily. No fitness landscape analysis was provided for the underlying fitness landscape of the controllers search space. As such, the literature remains largely inconclusive as to which ANN architecture provides the most efficient and effective space for searching the range of possible controllers through evolutionary methods. This represents the motivation for this paper where we compare the search space for four different types of ANN architecture for controller evolution through an information-theoretic analysis of the fitness landscape associated with each type of architecture.     

A fault-tolerant approach to robot teams

As the applications of mobile robotics evolve it has become increasingly less practical for researchers to design custom hardware and control systems for each problem. This paper presents a new approach to control system design in order to look beyond end-of-lifecycle performance, and consider control system structure, flexibility, and extensibility. Towards these ends the Control ad libitum philosophy was proposed, stating that to make significant progress in the real-world application of mobile robot teams the control system must be structured such that teams can be formed in real-time from diverse components. The Control ad libitum philosophy was applied to the design of the HAA (Host, Avatar, Agent) architecture: a modular hierarchical framework built with provably correct distributed algorithms. A control system for mapping, exploration, and foraging was developed using the HAA architecture and evaluated in three experiments. First, the basic functionality of the HAA architecture was studied, specifically the ability to: (a) dynamically form the control system, (b) dynamically form the robot team, (c) dynamically form the processing network, and (d) handle heterogeneous teams and allocate robots between tasks based on their capabilities. Secondly, the control system was tested with different rates of software failure and was able to successfully complete its tasks even when each module was set to fail every 0.5–1.5 min. Thirdly, the control system was subjected to concurrent software and hardware failures, and was still able to complete a foraging task in a 216 m² environment.     

基于遗传算法和直接搜索策略的PID整定研究

该文在详细分析遗传算法和直接搜索法优缺点的基础上提出了一种基于遗传算法和直接搜索策略的混合优化算法。该算法既具有遗传算法的全局寻优能力,又具有直接搜索法的局部寻优能力。可大大提高寻优的精度和速度。该混合算法先用遗传算法对给定区域进行全局的粗略搜索,然后用直接搜索法对其中部分较优个体进行局部的精细搜索。应用于PID自整定的仿真实验表明：该算法可节约绝大部分的进化代数,极大地提高寻优的速度,同时,PID整定的参数一致性好,具有比遗传退火策略更一致的寻优精度。     

遗传算法中遗传算子的启发式构造策略

遗传算法是影响遗传算法搜索性能的重要因素,本文研究交配算子与其搜索子空间的关系,提出了设计良好算子的指导性原则,并构造出一种启发式交配算子。