Cue integration by bumblebees during navigation in uncertain environments with multiplegoal options: Behavioural analysis in virtual reality and computational modelling

在不确定和混乱的环境中，许多动物可以准确地在它们的家和行为相关的地点（如食物来源）之间导航。我们的项目将通过行为和计算方法分析这些非凡能力背后的机制。更具体地说，我们将研究大黄蜂寻找目标和决策的机制，因为它们需要在环境中的许多其他物体中识别有价值的食物来源。我们将在定制设计的虚拟现实（VR）嵌入式轨迹球设置中使用视觉和嗅觉刺激对行走的大黄蜂进行行为实验，并密切结合潜在控制机制的计算建模。大黄蜂是这类行为研究的一个既定模型系统，也是它们导航系统的一部分。我们的项目旨在调查目标寻找行为的几个方面，包括：最初没有经验的蜜蜂是如何在许多其他物体中第一次发现一个有价值的目标的；蜜蜂是如何通过学习不同的环境线索来帮助定位和识别有利可图的食物来源，从而获得经验并改进他们的目标寻找策略的；蜜蜂是如何在具有不同奖励价值的多个目标选择的混乱环境中做出决定的；蜜蜂是如何学会区分可靠的环境特征和那些可能变化的环境特征的。我们项目的发现将为动物的目标导向行为和决策机制提供新的见解。通过实现一个模块化的模拟代理，该代理接收在VR设置的行为实验中提供给动物的感官输入，并产生类似蜜蜂的行为作为响应，我们将测试形式化的计算模型是否足以定性地再现所观察到的大黄蜂行为。这种使用VR技术和计算建模的实验分析的复杂互动将帮助我们更全面地了解昆虫寻找目标的复杂和多方面的过程。

Many animals can accurately navigate between their home and behaviourally relevant locations, such as food sources, in uncertain and cluttered environments. Our project will analyse by behavioural and computational approaches the mechanisms underlying these remarkable abilities. More specifically, we will investigate the mechanisms involved in goal finding and decision making in bumblebees as they need to identify rewarding food sources among the many other objects in their environment. We will conduct behavioural experiments on walking bumblebees using visual and olfactory stimuli in a custom-designed virtual reality (VR) embedded trackball setup in close combination with computational modelling of the underlying control mechanisms. Bumblebees are an established model system for this type of behavioural research and parts of their navigation system. Our project aims to investigate several aspects of goal-finding behaviour, including: how initially inexperienced bees discover a rewarding goal among many other objects for the first time, how bees gain experience and improve their goal finding strategies by learning different environmental cues that aid in locating and recognizing profitable food sources, how bees make decisions in cluttered environments with multiple goal options of different reward value, how bees learn to distinguish reliable environmental features from those that can change. Our project's findings will provide new insights into the mechanisms involved in goal-directed behaviour and decision-making in animals. By implementing a modular simulated agent that receives the sensory input provided to the animals in the behavioural experiments in the VR setup and generates bee-like behaviour in response, we will test if the formalized computational model is sufficient to qualitatively reproduce the observed behaviour of the bumblebees. This intricate interaction of experimental analysis using VR technology and computational modelling will help us gain a more comprehensive understanding of the complex and multi-faceted processes of goal finding in insects.