Swarm exploration and Communications: Integration by probabilistic learning (SCIL)

分布式人工智能(DAI)是人工智能中处理人工智能计算并发性的部分，即处理多智能体系统中知识和行动的分布和协调问题。涉及多智能体的探索问题--群体探索--就是DAI的很好的例证。群体探测是指在群体中对传感数据进行分布式处理，以重建未知的物理、化学或其他感兴趣的过程。它结合了分布式传感、优化(智能)信息收集和代理移动/动作协调的方法。因此，通信始终是群体探索的组成部分。群体探索通常认为可靠和无错误的通信。然而，通信系统确实增加了交换的信息的不确定性。例如，在为代理预测新的采样位置时，需要考虑通信的不确定性，因为导致严重通信退化的位置对于分布式信息处理/探索目的将是无用的。同样，通信系统旨在实现测量或处理结果的无差错传输，但它们既不知道它们与学习整个探索过程的相关性，也不知道要传输的数据的可信度。这个研究项目的主要目标是开发一个框架和方法，以及设计、调查和概念验证算法，将探索和交流“紧密”结合在一起。为此，我们利用概率学习的工具，使探索“通信感知”，通信“探索感知”，从而在DAI意义上实现多主体网络中的知识和行为的协调。考虑到这一点，这项工作中提出的研究旨在解决未来分布式传感和数据处理平台-传感器网络或由多个代理组成的移动机器人群-收集、通信和处理空间分布的传感器数据的挑战。

Distributed Artificial Intelligence (DAI) is the part of AI that deals withthe concurrency of AI computations, i.e. with problems of distributionand coordination of knowledge and action in multi-agent systems.Exploration problems involving multiple agents – swarm exploration –exemplify well DAI. Swarm exploration is concerned with distributedprocessing within the swarm of sensed data for reconstructing anunknown physical, chemical or other process of interest. Itincorporates methods for distributed sensing, optimized (intelligent)information gathering and agent movement/action coordination.Communication is therefore always an integral part of a swarmexploration. Swarm exploration often considers reliable and error-freecommunications. However, communication systems do adduncertainty to exchanged information. For instance, communicationuncertainty needs to be taken into account when predicting newsampling positions for agents as locations causing severecommunication degradation will be useless for distributed informationprocessing/exploration purposes. Likewise, communication systemsare designed to aim for error-free transmission of measurements orprocessing results, but they are neither aware of their relevance forlearning the entire explored process nor of the confidence in the datato be transmitted. The main goal of this research project is thedevelopment of a framework and methodology as well as the design,investigation and proof-of-concept demonstration of algorithms of a“tight” integration of exploration and communication. To this end, wemake exploration “communication-aware” and communications“exploration-aware” by using tools of probabilistic learning, thusenabling the coordination of knowledge and action in multi-agentnetworks in the sense of DAI. With this in mind, the researchproposed in this work addresses challenges in development of futuredistributed sensing and data processing platforms – sensor networksor mobile robotic swarms consisting of multiple agents – that cancollect, communicate, and process spatially distributed sensor data.