Search

Several methods have already been proposed to automatically design control software for robot swarms by assembling predefined modules. Yet, so far, the modules on which these methods operate have always been defined manually in a process that is time consuming, requires domain knowledge, and must be performed by an expert. Motivated by the goal of automatizing the definition of these modules, we propose an approach in which repertoires of modules, in the form of neural networks, are automatically generated via a quality-diversity evolutionary algorithm. To illustrate the proposal, we introduce Nata, a novel approach belonging to the AutoMoDe family. Nata automatically generates probabilistic finite-state machines in which states are selected from a repertoire of neural networks, and transition conditions are selected from a set of rules based on the sensory capabilities of the robotic platform considered. Both the repertoire of neural networks and the set of transition rules are automatically generated a priori, once and for all, in a mission-agnostic way. We study Nata on three missions, both in simulation and with real robots. Nata is the first modular automatic design method that assembles modules that were themselves generated automatically. [Display omitted] • Automatic modular methods for designing robot swarms can be automatically generated. • Nata selects and combines neural networks generated a priori via novelty search. • Nata outperforms the classical neuroevolutionary approach. [ABSTRACT FROM AUTHOR]

Abstract Recent works in evolutionary robotics have shown the viability of evolution driven by behavioural novelty and diversity. These evolutionary approaches have been successfully used to generate repertoires of diverse and high-quality behaviours, instead of driving evolution towards a single, task-specific solution. Having repertoires of behaviours can enable new forms of robotic control, in which high-level controllers continually decide which behaviour to execute. To date, however, only the use of repertoires of open-loop locomotion primitives has been studied. We propose EvoRBC-II, an approach that enables the evolution of repertoires composed of general closed-loop behaviours, that can respond to the robot's sensory inputs. The evolved repertoire is then used as a basis to evolve a transparent higher-level controller that decides when and which behaviours of the repertoire to execute. Relying on experiments in a simulated domain, we show that the evolved repertoires are composed of highly diverse and useful behaviours. The same repertoire contains sufficiently diverse behaviours to solve a wide range of tasks, and the EvoRBC-II approach can yield a performance that is comparable to the standard tabula-rasa evolution. EvoRBC-II enables automatic generation of hierarchical control through a two-step evolutionary process, thus opening doors for the further exploration of the advantages that can be brought by hierarchical control. [ABSTRACT FROM AUTHOR]

• Automatic methods can design spatially-organizing behaviors for robot swarms. • AutoMoDe-Mate is a design method specialized in spatial organization problems. • AutoMoDe-Mate includes a module that enables robot pattern formation. • Experiments with physical robots show effectiveness in coverage and networking. • AutoMoDe-Mate outperforms baseline modular and neuro-evolutionary methods. [Display omitted] We present Mate, an automatic off-line design method specialized in the design of spatially-organizing behaviors for robot swarms. Mate belongs to the family of modular methods known as AutoMoDe. We introduce Mate to study the automatic design of collective behaviors for missions in which the swarm is subject to spatial distribution constrains. In this paper, we produce control software for three missions with specifications related to the distribution of the swarm in the environment. We conduct experiments in simulation and with a swarm of 20 e-puck robots. Alongside Mate, we also conduct experiments with two other automatic design methods: Chocolate—a state-of-the-art instance of AutoMoDe; and EvoSpace—a method based on neuro-evolution. Early studies conducted with existing modular design methods have shown their limitations in the design of spatially-organizing behaviors for robots that operate under spatial constrains. By introducing a specialized method like Mate, we expect to overcome these limitations. The aggregate results of our experiments show that Mate performs significantly better than Chocolate and EvoSpace in the missions we consider. [ABSTRACT FROM AUTHOR]

With the increasing task complexity and environmental uncertainty, it is hard to achieve adaptability and robustness by manual design methods for multi-robot cooperation tasks. Automatic synthesis approaches with trial and error mechanisms are getting more and more attention. By encoding the strategies to be designed as "ideas", the newly proposed Brain Storm Robotics (BSR) framework can obtain the sufficiently good solutions for particular tasks after a series of operations on the ideas. However, the original BSR only shows designing the rule base for a fuzzy controller. This paper proposes an automatic design approach for neural network-based strategies for robotic swarms with the BSR framework to realize cooperative behaviors. Two design cases are studied: one is the direct strategy search for a swarm aggregation behavior; the other is synthesizing a backpropagation neural network-based controller for coordinated formation control, which has both evolution and learning characteristics. The results show that the proposed method can automatically find control strategies with scalability for multi-robot cooperation, which has the potential for further development. • The BSR framework can synthesize neural network-based strategies automatically. • It can synthesize direct control strategies and parameters for learning mechanisms. • Two examples are given: aggregation and a formation control for robotic swarms. [ABSTRACT FROM AUTHOR]

Evolutionary robotics employs unconventional techniques to continuously evolve controllers for robots, based on their fitness values. In most cases, a parent controller is subjected to mutation to evolve its offspring. If the offspring performs better than the parent, the former is made to replace the latter. This essentially results in a major loss in information learned by the parents over generations. One simple workaround to circumvent this problem is to maintain a Hall of Fame (HoF) comprising the best parent controllers for use in future generations. In embodied evolutionary robotic scenarios, caching a large number of controllers in an HoF would result in increased computational overheads while selecting the best out of them, resulting in a drastic reduction in performance. With no means to find an upper limit to the number of controllers that can populate an HoF a priori , devising a technique to dynamically regulate this population is imperative. In this work, a novel method to evict the non-performing controllers within an HoF based on the dynamics of the system, is proposed. We describe the evolution of such controllers using genetic operators that eventually form an Idiotypic Network. A constantly varying Resource , associated with each controller together with its concentration in the Idiotypic Network, helps decide its eviction from the HoF. Experiments performed using simulations and also a real robot indicate a marked improvement in the learning process due to the dynamic eviction policy. [ABSTRACT FROM AUTHOR]

Evolutionary robots can achieve certain goals via evolutionary algorithms without specifying all the detailed actions. The robot interacts with the environment and receives natural feedback as the fitness of its goal. In this paper, we study two biomorphic robots and observe how the behavior can evolve with least computation resources and test the possibility of self-adjustment on an unexpected motor failure of one robot leg. The feedback is collected via a supersonic sensor and the evolutionary algorithm is a modified low cost genetic algorithm. In order to minimize the cost of evolution, we control the number of the population of each generation to a very low number. We find that the biomorphic robots can move faster with the motion sequence generated by the evolutionary algorithm than with the motion sequence assigned by human programmers. The structures of robots, in this case, and the number of legs might affect the speed of convergence. Our experiment result shows that the robot can adjust itself with this online evolving approach to deal with an unexpected motor fault. That is, evolutionary robot can overcome an unexpected situation without human intervention. [ABSTRACT FROM AUTHOR]

Physics simulation and character control are two important issues in computer games. In this paper, we propose two games which are tailored for investigating some aspects of these two issues. We study on the applications of neural network and the genetic algorithm techniques for building the controllers and the controllers should be able to finish the specific tasks in the two games. The goal of the first game is that the controller can shoot a ball so that the ball collides with the other two balls one after another. The challenge of this game is that the ball should be shot from the proper position and the goal is achieved every time. The second game is a duel game and two virtual characters are controlled to fight with each other. We develop a method for verifying whether or not the skill power of the two virtual characters is balanced. The controllers of both games are evolved based on neural network and genetic algorithm in an unsupervised learning manner. We perform a comprehensive study on the performance and weaknesses of the controllers. [ABSTRACT FROM AUTHOR]

Ad hoc mreže specifične su po svojim karakteristikama (necentraliziranost, samoorganiziranje I višeskokovnost). Zbog toga usmjerni protokoli u ad hoc mrežama moraju biti prilagođeni pojedinim primjenama reže i zadovolja- vati glavne zahtjeve -- propusnosti, skalabilnost i sigurnost. U radu je prikazan pregled sigurnosti usmjeravanja u ad hoc mrežama. Navedena su svojstva ad hoc mreža I izazovi razvoja protokola usmjeravanja u njima. Opisani su glavni sigurnosni problemi u usmjeravanju i metode ispitivanja sigurnosti usmjernih protokola. [ABSTRACT FROM AUTHOR]

This paper describes a new approach in using multi-objective evolutionary algorithms in evolving the neural network that acts as a controller for the phototaxis and radio frequency localization behaviors of a virtual Khepera robot simulated in a 3D, physics-based environment. The Pareto-frontier Differential Evolution (PDE) algorithm is utilized to generate the Pareto optimal sets through a 3-layer feed-forward artificial neural network that optimize the conflicting objectives of robot behavior and network complexity, where the two different types of robot behaviors are phototaxis and RF-localization, respectively. Thus, there are two fitness functions proposed in this study. The testing results showed the robot was able to track the light source and also home-in towards the RF-signal source successfully. Furthermore, three additional testing results have been incorporated from the robustness perspective: different robot localizations, inclusion of two obstacles, and moving signal source experiments, respectively. The testing results also showed that the robot was robust to these different environments used during the testing phases. Hence, the results demonstrated that the utilization of the evolutionary multi-objective approach in evolutionary robotics can be practically used to generate controllers for phototaxis and RF-localization behaviors in autonomous mobile robots. [ABSTRACT FROM AUTHOR]

• We address time efficiency further to traditional data efficiency to evaluate generate-and-test style optimization algorithms and a precise way to measure time efficiency by gain per time unit. • To the time-efficient search algorithm, we propose a new algorithm, Bayesian-Evolutionary Algorithm that combines time efficiency and data efficiency by combining Bayesian and evolutionary optimization, and a method to select and transfer knowledge from the Bayesian to the evolutionary component. • We demonstrate the merit of the BEA as a learning method on well-known objective functions and tasks in evolutionary robotics. We investigate its performance regarding fitness, computation time, and time efficiency (gain per time unit) and implement a demonstration of the best controllers learned by the BEA on real robots. Not all generate-and-test search algorithms are created equal. Bayesian Optimization (BO) invests a lot of computation time to generate the candidate solution that best balances the predicted value and the uncertainty given all previous data, taking increasingly more time as the number of evaluations performed grows. Evolutionary Algorithms (EA) on the other hand rely on search heuristics that typically do not depend on all previous data and can be done in constant time. Both BO and EA community typically assess their performance as a function of the number of evaluations, i.e., data efficiency. However, this is unfair once we start to compare the efficiency of these classes of algorithms, as the overhead times to generate candidate solutions are significantly different. We suggest to measure the efficiency of generate-and-test search algorithms as the expected gain in the objective value per unit of computation time spent, i.e., time efficiency. To the time-efficient search algorithm, we therefore propose a new algorithm, a combination of BO and an EA, BEA for short, that starts with BO, then transfers knowledge to an EA, and subsequently runs the EA. We compare the BEA with BO, the EA, Differential Evolution (DE), and Particle Swarm Optimization (PSO). The results show that BEA outperforms BO, the EA, DE and PSO in terms of time efficiency, and ultimately leads to better performance on well-known benchmark objective functions with many local optima. Moreover, we test BEA, BO, and the EA on nine test cases of robot learning problems and here again we find that BEA outperforms the other algorithms. [ABSTRACT FROM AUTHOR]

Objective: Robotic telemanipulation systems have been introduced recently to enhance the surgeon''s dexterity and visualization in endoscopic surgery and thus facilitate refined dissection, suturing, and knot tying. This study examined the technical feasibility of performing safe and efficient robot-assisted, hand-sewn endoscopic airway reconstruction in a rabbit model. Methods: A total of 18 tracheal anastomoses were performed in rabbits, with 6 anastomoses performed endoscopically using the da Vinci Surgical system (Robot group), 6 anastomoses performed using traditional thoracoscopic surgical instruments (VATS group), and 6 anastomoses performed using open surgical instruments (Open group). Anastomosis time and complications were recorded. The effectiveness of anastomoses was evaluated by postoperative observation of all rabbits for 8 weeks and measurement of anastomotic stricture and pathologic findings. These parameters and anastomosis time were compared between groups. Results: In all cases in the Robot group, the procedure was completed endoscopically. No perioperative or postoperative complications were encountered. Mean procedure time of the Robot group was 14.1 ± 2.6 minutes (mean ± standard deviation). Anastomosis time in the Robot group was significantly shorter than in the VATS group (33.5 ± 5.2 minutes, P = .0039) and was not significantly different in the Open group (11.4 ± 2.3 minutes, P = .1282). All anastomoses in the Robot group remained mechanically intact, and all parameters were comparable with those of the open surgery group. Conclusions: The technical feasibility of performing safe and efficient robot-assisted endoscopic airway reconstruction was repeatedly demonstrated in a rabbit model. Robotic assistance significantly improved the time associated with and the quality of endoscopic airway reconstruction. [Copyright &y& Elsevier]

Abstract: Evolutionary Robotics (ER) is one of promising approaches to design robot controllers which essentially have complicated and/or complex properties. In most ER research, the sensory–motor mappings of robots are represented as artificial neural networks, and their connection weights (and sometimes the structure of the networks) can be optimized in the parameter spaces by using evolutionary computation. However, generally, the evolved neural controllers could be fragile in unexperienced environments, especially in real worlds, because the evolutionary optimization processes would be executed in idealized simulators. This is known as the gap problem between the simulated and real worlds. To overcome this, the author focused on evolving an on-line learning ability instead of weight parameters in a simulated environment. According to recent biological findings, actually, the kinds of on-line adaptation abilities can be found in real nervous systems of insects and crustaceans, and it is also known that a variety of neuromodulators (NMs) play crucial roles to regulate the network characteristics (i.e. activating/blocking/changing of synaptic connections). Based on this, a neuromodulatory neural network model was proposed and it was utilized as a mobile robot controller. In the paper, the detail behavior analysis of the evolved neuromodulatory neural network is also discussed. [Copyright &y& Elsevier]

Abstract: This paper describes an evolutionary way to acquire behaviors of a mobile robot for recognizing environments. We have proposed Action-based Environment Modeling (AEM) approach for a simple mobile robot to recognize environments. In AEM, a behavior-based mobile robot acts in each environment and action sequences are obtained. The action sequences are transformed into vectors characterizing the environments, and the robot identifies the environments with similarity between the vectors. The suitable behaviors like wall-following for AEM have been designed by a human. However the design is very difficult for him/her because the search space is huge and intuitive understanding is hard. Thus we apply evolutionary robotics approach to design of such behaviors using genetic algorithm and make simulations in which a robot recognizes the environments with different structures. As results, we find out suitable behaviors are learned even for environments in which human hardly designs them, and the learned behaviors are more efficient than hand-coded ones. [Copyright &y& Elsevier]

The vision behind this paper looks ahead to evolutionary robot systems where morphologies and controllers are evolved together and 'newborn' robots undergo a learning process to optimize their inherited brain for the inherited body. The specific problem we address is learning controllers for the task of directed locomotion in evolvable modular robots. To this end, we present a test suite of robots with different shapes and sizes and compare two learning algorithms, Bayesian optimization and HyperNEAT. The experiments in simulation show that both methods obtain good controllers, but Bayesian optimization is more effective and sample efficient. We validate the best learned controllers by constructing three robots from the test suite in the real world and observe their fitness and actual trajectories. The obtained results indicate a reality gap, but overall the trajectories are adequate and follow the target directions successfully. • We achieve the task of directed locomotion in evolvable modular robots rather than a typical undirected gait learning. • We investigate the generic methods that allow modular robots with arbitrary shapes and sizes for directed locomotion. • We validate the learned controllers for directed locomotion on real robots and reflect on the reality gap. [ABSTRACT FROM AUTHOR]

Abstract: Interdisciplinary research in the 21st century is characterized by bidirectional flows: one domain provides inspiration to another, which, after an advance, provides inspiration back to the donating domain. In this abstract I outline three such flows between the domains of evolutionary biology and robotics. First, biological evolution shapes all aspects of an organism''s body and brain simultaneously. This led to work in which artificial evolution optimizes the morphology and neural control of robots such that they perform increasingly sophisticated tasks. Second, evolution causes change over evolutionary time, but also over the lifetime of the organism. This led to work in which virtual robots change body plans as they evolve to perform more complex tasks, but their bodies also change as they perform those tasks. Finally, evolution always works on populations. This led to work in which populations of humans collaborate and compete to evolve increasingly sophisticated robots. [Copyright &y& Elsevier]