Swarm Intelligence最新文献

Grammatical evolution can be used to learn bio-inspired solutions to many distributed multiagent tasks, but the programs learned by the agents often need to be resilient to perturbations in the world. Biological inspiration from bacteria suggests that ongoing evolution can enable resilience, but traditional grammatical evolution algorithms learn too slowly to mimic rapid evolution because they utilize only vertical, parent-to-child genetic variation. The BeTr-GEESE grammatical evolution algorithm presented in this paper creates agents that use both vertical and lateral gene transfer to rapidly learn programs that perform one step in a multi-step problem even though the programs cannot perform all required subtasks. This paper shows that BeTr-GEESE can use online evolution to produce resilient collective behaviors on two goal-oriented spatial tasks, foraging and nest maintenance, in the presence of different types of perturbation. The paper then explores when and why BeTr-GEESE succeeds, emphasizing two potentially generalizable properties: modularity and locality. Modular programs enable real-time lateral transfer, leading to resilience. Locality means that the appropriate phenotypic behaviors are local to specific regions of the world (spatial locality) and that recently useful behaviors are likely to be useful again shortly (temporal locality). Finally, the paper modifies BeTr-GEESE to perform behavior fusion across multiple modular behaviors using activator and repressed conditions so that a fixed (non-evolving) population of heterogeneous agents is resilient to perturbations.

Coordination of local and global aerial traffic has become a legal and technological bottleneck as the number of unmanned vehicles in the common airspace continues to grow. To meet this challenge, automation and decentralization of control is an unavoidable requirement. In this paper, we present a solution that enables self-organization of cooperating autonomous agents into an effective traffic flow state in which the common aerial coordination task—filled with conflicts—is resolved. Using realistic simulations, we show that our algorithm is safe, efficient, and scalable regarding the number of drones and their speed range, while it can also handle heterogeneous agents and even pairwise priorities between them. The algorithm works in any sparse or dense traffic scenario in two dimensions and can be made increasingly efficient by a layered flight space structure in three dimensions. To support the feasibility of our solution, we show stable traffic simulations with up to 5000 agents, and experimentally demonstrate coordinated aerial traffic of 100 autonomous drones within a 250 m wide circular area.

Programmable and self-assembled magnetic pillars are essential to expanding the application domain of magnetic microparticle collectives. Typically, the collective behavior of self-assembled magnetic pillars is carried out by generating uniform and time-varying magnetic fields. However, magnetic field-shaping capabilities employing non-uniform fields have not been explored for magnetic pillars. In this study, we generate non-uniform magnetic fields using a nine-coil electromagnetic system to achieve object manipulation, upstream/downstream locomotion, and independent actuation. We begin analyzing the static magnetic self-assembly of reduced iron microparticles and experimentally derive the average dimensions (height and diameter) of the resulting pillars. Subsequently, we delve into the collective dynamic response under non-uniform and time-varying magnetic fields, unveiling four distinct modalities. In order to demonstrate the versatility of our approach, we extend our study to the two-dimensional manipulation of a millimeter-sized glass bead using a precessing magnetic field describing a Lissajous curve. Moreover, we showcase the ability of magnetic pillars to adapt to confined and dynamic conditions within fluidic tubes. We finally present a noteworthy case where the nine-coil electromagnetic system independently actuates two clusters of magnetic pillars. Our study shows the potential of using non-uniform magnetic fields to actuate self-assembled magnetic pillars, enabling morphology reconfiguration capabilities, object manipulation, locomotion, and independent actuation.

Applications, such as military and disaster response, can benefit from robotic collectives’ ability to perform multiple cooperative tasks (e.g., surveillance, damage assessments) efficiently across a large spatial area. Coalition formation algorithms can potentially facilitate collective robots’ assignment to appropriate task teams; however, most coalition formation algorithms were designed for smaller multiple robot systems (i.e., 2–50 robots). Collectives’ scale and domain-relevant constraints (i.e., distribution, near real-time, minimal communication) make coalition formation more challenging. This manuscript identifies the challenges inherent to designing coalition formation algorithms for very large collectives (e.g., 1000 robots). A survey of multiple robot coalition formation algorithms finds that most are unable to transfer directly to collectives, due to the identified system differences; however, auctions and hedonic games may be the most transferable. A simulation-based evaluation of five total algorithms from two combinatorial auction families and one hedonic game family, applied to homogeneous and heterogeneous collectives, demonstrates that there are collective compositions for which no evaluated algorithm is viable; however, the experimental results and literature survey suggest paths forward.

Social learning is a collective approach to decentralised decision-making and is comprised of two processes; evidence updating and belief fusion. In this paper we propose a social learning model in which agents’ beliefs are represented by a set of possible states, and where the evidence collected can vary in its level of imprecision. We investigate this model using multi-agent and multi-robot simulations and demonstrate that it is robust to imprecise evidence. Our results also show that certain kinds of imprecise evidence can enhance the efficacy of the learning process in the presence of sensor errors.

Colonies of ants can complete complex tasks without the need for centralised control as a result of interactions between individuals and their environment. Particularly remarkable is the process of path selection between the nest and food sources that is essential for successful foraging. We have designed a stochastic model of ant foraging in the absence of direct communication. The motion of ants is governed by two components - a random change in direction of motion that improves ability to explore the environment, and a non-random global indirect interaction component based on pheromone signalling. Our model couples individual-based off-lattice ant simulations with an on-lattice characterisation of the pheromone diffusion. Using numerical simulations we have tested three pheromone-based model alternatives: (1) a single pheromone laid on the way toward the food source and on the way back to the nest; (2) single pheromone laid on the way toward the food source and an internal imperfect compass to navigate toward the nest; (3) two different pheromones, each used for one direction. We have studied the model behaviour in different parameter regimes and tested the ability of our simulated ants to form trails and adapt to environmental changes. The simulated ants behaviour reproduced the behaviours observed experimentally. Furthermore we tested two biological hypotheses on the impact of the quality of the food source on the dynamics. We found that increasing pheromone deposition for the richer food sources has a larger impact on the dynamics than elevation of the ant recruitment level for the richer food sources.

Abstract

An emerging challenge in swarm shepherding research is to design effective and efficient artificial intelligence algorithms that maintain simplicity in their decision models, whilst increasing the swarm’s abilities to operate in diverse contexts. We propose a methodology to design a context-aware swarm control intelligent agent (shepherd). We first use swarm metrics to recognise the type of swarm that the shepherd interacts with, then select a suitable parameterisation from its behavioural library for that particular swarm type. The design principle of our methodology is to increase the situation awareness (i.e. contents) of the control agent without sacrificing the low computational cost necessary for efficient swarm control. We demonstrate successful shepherding in both homogeneous and heterogeneous swarms.

Biologically inspired collective decision-making algorithms show promise for implementing spatially distributed searching tasks with robotic systems. One example is the best-of-N problem in which a collective must search an environment for an unknown number of sites and select the best option. Real-world robotic deployments must achieve acceptable success rates and execution times across a wide variety of environmental conditions, a property known as resilience. Existing experiments for the best-of-N problem have not explicitly examined how the site layout affects a collective’s performance and resilience. Two novel resilience metrics are used to compare algorithmic performance and resilience between evenly distributed, obstructed, or unobstructed uneven site configurations. Obstructing the highest valued site negatively affected selection accuracy for both algorithms, while uneven site distribution had no effect on either algorithm’s resilience. The results also illuminate the distinction between absolute resilience as measured against an objective standard, and relative resilience used to compare an algorithm’s performance across different operating conditions.

One of the main challenges in swarm robotics is to achieve robust and scalable flocking, such that large numbers of robots can move together in a coordinated and cohesive manner while avoiding obstacles or threats. Flocking models in swarm robotic systems typically use reactive behaviors, such as cohesion, alignment, and avoidance. The use of potential fields has enabled the derivation of reactive control laws using obstacles and neighboring robots as sources of force for flocking. However, reactive behaviors, especially when a multitude of them are simultaneously active, as in the case of flocking, are prone to cause collisions or inefficient motion within the flock due to its short-sighted approach. Approaches that aimed to generate smoother and optimum flocking, such as the use of model predictive control, would either require centralized coordination, or distributed coordination which requires low-latency and high-bandwidth communication requirements within the swarm as well as high computational resources. In this paper, we present a predictive search model that can generate smooth and safe flocking of robotic swarms in the presence of obstacles by taking into account the predicted states of other robots in a computationally efficient way. We tested the proposed model in environments with static and dynamic obstacles and compared its performance with a potential field flocking model in simulation. The results show that the predictive search model can generate smoother and faster flocking in swarm robotic systems in the presence of static and dynamic obstacles. Furthermore, we tested the predictive search model with different numbers of robots in environments with static obstacles in simulations and demonstrated that it is scalable to large swarm sizes. The performance of the predictive search model is also validated on a swarm of six quadcopters indoors in the presence of static and dynamic obstacles.

Particle swarm optimization (PSO) performance is sensitive to the control parameter values used, but tuning of control parameters for the problem at hand is computationally expensive. Self-adaptive particle swarm optimization (SAPSO) algorithms attempt to adjust control parameters during the optimization process, ideally without introducing additional control parameters to which the performance is sensitive. This paper proposes a belief space (BS) approach, borrowed from cultural algorithms (CAs), towards development of a SAPSO. The resulting BS-SAPSO utilizes a belief space to direct the search for optimal control parameter values by excluding non-promising configurations from the control parameter space. The resulting BS-SAPSO achieves an improvement in performance of 3–55% above the various baselines, based on the solution quality of the objective function values achieved on the functions tested.