Paulo Urbano

Evolution of swarm robotics systems with novelty search

Novelty search is a recent artificial evolution technique that challenges traditional evolutionary approaches. In novelty search, solutions are rewarded based on their novelty, rather than their quality with respect to a predefined objective. The lack of a predefined objective precludes premature convergence caused by a deceptive fitness function. In this paper, we apply novelty search combined with NEAT to the evolution of neural controllers for homogeneous swarms of robots. Our empirical study is conducted in simulation, and we use a common swarm robotics task—aggregation, and a more challenging task—sharing of an energy recharging station. Our results show that novelty search is unaffected by deception, is notably effective in bootstrapping evolution, can find solutions with lower complexity than fitness-based evolution, and can find a broad diversity of solutions for the same task. Even in non-deceptive setups, novelty search achieves solution qualities similar to those obtained in traditional fitness-based evolution. Our study also encompasses variants of novelty search that work in concert with fitness-based evolution to combine the exploratory character of novelty search with the exploitatory character of objective-based evolution. We show that these variants can further improve the performance of novelty search. Overall, our study shows that novelty search is a promising alternative for the evolution of controllers for robotic swarms.

odNEAT: An Algorithm for Distributed Online, Onboard Evolution of Robot Behaviours

We propose and evaluate a novel approach called Online Distributed NeuroEvolution of Augmenting Topologies (odNEAT). odNEAT is a completely distributed evolutionary algorithm for online learning in groups of embodied agents such as robots. While previous approaches to online distributed evolution of neural controllers have been limited to the optimisation of weights, odNEAT evolves both weights and network topology. We demonstrate odNEAT through a series of simulation-based experiments in which a group of e-puck-like robots must perform an aggregation task. Our results show that robots are capable of evolving effective aggregation strategies and that sustainable behaviours evolve quickly. We show that odNEAT approximates the performance of rtNEAT, a similar but centralised method. We also analyse the contribution of each algorithmic component on the performance through a series of ablation studies.

Tax compliance in a simulated heterogeneous multi-agent society

We consider an individualised approach to agent behaviour in an application to the classical economic problem of tax compliance. Most economic theories consider homogeneous representative agent utilitarian approaches to explain the decision of complying or not with tax payment. However, a heterogeneous and individualised account of decision can be considered to explain certain apparently irrational behaviours. Ideas such as trust and peer perception may have a key influence in individual decisions, and thus transform the global results for society. In this paper, we apply the agent view of rationality to economic decisions and define a territory to be explored by agent technology and social simulations. We conclude that the multi-agent view can provide powerful results which might lead to significant economic policy implications.

Odneat: An algorithm for decentralised online evolution of robotic controllers

Online evolution gives robots the capacity to learn new tasks and to adapt to changing environmental conditions during task execution. Previous approaches to online evolution of neural controllers are typically limited to the optimisation of weights in networks with a prespecified, fixed topology. In this article, we propose a novel approach to online learning in groups of autonomous robots called odNEAT. odNEAT is a distributed and decentralised neuroevolution algorithm that evolves both weights and network topology. We demonstrate odNEAT in three multirobot tasks: aggregation, integrated navigation and obstacle avoidance, and phototaxis. Results show that odNEAT approximates the performance of rtNEAT, an efficient centralised method, and outperforms IM-(), a decentralised neuroevolution algorithm. Compared with rtNEAT and IM-(), odNEAT’s evolutionary dynamics lead to the synthesis of less complex neural controllers with superior generalisation capabilities. We show that robots executing odNEAT can display a high degree of fault tolerance as they are able to adapt and learn new behaviours in the presence of faults. We conclude with a series of ablation studies to analyse the impact of each algorithmic component on performance.

Progressive Minimal Criteria Novelty Search

We propose progressive minimal criteria novelty search (PMCNS), which is an extension of minimal criteria novelty search. In PMCNS, we combine the respective benefits of novelty search and fitness-based evolution by letting novelty search freely explore new regions of behaviour space as long as the solutions meet a progressively stricter fitness criterion. We evaluate the performance of our approach in the evolution of neurocontrollers for a swarm of robots in a coordination task where robots must share a single charging station. The robots can only survive by periodically recharging their batteries. We compare the performance of PMCNS with (i) minimal criteria novelty search, (ii) pure novelty search, (iii) pure fitness-based evolution, and (iv) with evolutionary search based on a linear blend of novelty and fitness. Our results show that PMCNS outperforms all four approaches. Finally, we analyse how different parameter setting in PMCNS influence the exploration of the behaviour space.

A Land Use Planning Ontology: LBCS

Abstract: Urban planning has a considerable impact on the economic performance of cities and on the quality of life of their populations. Efficiency at this level has been hampered by the lack of integrated tools to adequately describe urban space in order to formulate appropriate design solutions. This paper describes an ontology called LBCS-OWL2 specifically developed to overcome this flaw, based on the Land Based Classification Standards (LBCS), a comprehensive and detailed land use standard to describe the different dimensions of urban space. The goal is to provide semantic and computer-readable land use descriptions of geo-referenced spatial data. This will help to make programming strategies available to those involved in the urban development process. There are several advantages to transferring a land use standard to an OWL2 land use ontology: it is modular, it can be shared and reused, it can be extended and data consistency maintained, and it is ready for integration, thereby supporting the interoperability of different urban planning applications. This standard is used as a basic structure for the “City Information Modelling” (CIM) model developed within a larger research project called City Induction, which aims to develop a tool for urban planning and design.

Improving Grammatical Evolution in Santa Fe Trail using Novelty Search

Grammatical Evolution is an evolutionary algorithm that can evolve complete programs using a Backus Naur form grammar as a plug-in component to describe the output language. An important issue of Grammatical Evolution, and evolutionary computation in general, is the difficulty in dealing with deceptive problems and avoid premature convergence to local optima. Novelty search is a recent technique, which does not use the standard fitness function of evolutionary algorithms but follows the gradient of behavioral diversity. It has been successfully used for solving deceptive problems mainly in neuro-evolutionary robotics where it was originated. This work presents the first application of Novelty Search in Grammatical Evolution (as the search component of the later) and benchmarks this novel approach in a wellknown deceptive problem, the Santa Fe Trail. For the experiments, two grammars are used: one that defines a search space semantically equivalent to the original Santa Fe Trail problem as defined by Koza and a second one which were widely used in the Grammatical Evolution literature, but which defines a biased search space. The application of novelty search requires to characterize behavior, using behavior descriptors and compare descriptions using behavior similarity metrics. The conducted experiments compare the performance of standard Grammatical Evolution and its Novelty Search variation using four intuitive behavior descriptors. The experimental results demonstrate that Grammatical Evolution with Novelty Search outperforms the traditional fitness based Grammatical Evolution algorithm in the Santa Fe Trail problem demonstrating a higher success rates and better solutions in terms of the required steps.

The T. albipennis sand painting artists

In recent years, we have seen some artificial artistic work that has drawn inspiration from swarm societies, in particular ant societies. Ant paintings are abstract images corresponding to visualizations of the paths made by a group of virtual ants on a bi-dimensional space. The research on ant paintings has been focused around a stigmergic mechanism of interaction: the deposition of pheromones, largely used by ants. In an effort to further on the research on ant inspired artificial art, we introduce the T. albipennis sand painting artists, which draw direct inspiration from the ant species Temnothorax albipennis (formerly tuberointerruptus). These ants build simple circular walls, composed of grains of sand or fragments of stones, at a given distance from the central cluster of adult ants and brood. The brood and ants cluster function as a template, which combined with self-organization are responsible for the particular wall pattern formation. The T. albipennis artists are artificial twodimensional builders, starting from unorganized placement of virtual sand grains, they rearrange them, creating some interesting patterns composed of scattered pointillistic and imperfect circles, a colored moon-like landscape full of craters.

Force Versus Majority: A Comparison in Convention Emergence Efficiency

In open societies such as multi-agent systems, it is important that coordination among the several actors is achieved efficiently. One economical way of capturing that aspiration is consensus: social conventions and lexicons are good examples of coordinating systems, where uniformity promotes shared expectations of behavior and shared meanings. We are particularly interested in consensus that is achieved without any central control or ruling, through decentralized mechanisms that prove to be effective, efficient, and robust. The nature of interactions and also the nature of society configurations may promote or inhibit consensual emergence. Traditionally, preference to adopt the most seen choices (the majority option) has dominated the emergence convention research in multi-agents, being analyzed along different social topologies. Recently, we have introduced a different type of interaction, based on force, where force is not defined a priori but evolves dynamically. We compare the Majority class of choice update against Force based interactions, along three dimensions: types of encounters, rules of interaction and network topologies. Our experiments show that interactions based on Force are significantly more efficient (fewer encounters) for group decision making.

Adaptation of Robot Behaviour through Online Evolution and Neuromodulated Learning

We propose and evaluate a novel approach to the online synthesis of neural controllers for autonomous robots. We combine online evolution of weights and network topology with neuromodulated learning. We demonstrate our method through a series of simulation-based experiments in which an e-puck-like robot must perform a dynamic concurrent foraging task. In this task, scattered food items periodically change their nutritive value or become poisonous. Our results show that when neuromodulated learning is employed, neural controllers are synthesised faster than by evolution alone. We demonstrate that the online evolutionary process is capable of generating controllers well adapted to the periodic task changes. An analysis of the evolved networks shows that they are characterised by specialised modulatory neurons that exclusively regulate the output neurons.