Neuronal control of behaviour in complex sensory environments

大多数动物生活在复杂的环境中，这些环境往往包含着与周围环境相互矛盾的信息。因此，生存取决于神经系统收集和处理相关信息的能力，从而使动物产生适当的行为反应。为了理解行为的神经控制，我们必须首先了解动物如何在三维空间中定位，以及运动附件的运动学是如何由相关的肌肉活动驱动的，这些肌肉活动最终产生力量，使动物在其环境中移动。为了理解行为产生和改变的核心机制，我们必须研究神经系统如何检测环境信号并驱动适当的运动肌肉。虽然单个神经元可能适合于检测环境的特定方面，但神经元群的联合活动对于检测和破译复杂的感觉线索至关重要。从昆虫到人类等动物的常见检测方案表明，从环境中提取重要信息涉及类似的基本过程。通过识别神经元群从复杂刺激中检测相关线索的机制，也将有可能设计出更好的受生物启发的人工系统，这些系统具有适应性强、自我导向的导航能力。我的实验室使用昆虫飞行作为一个理想的模型系统来解决自然行为背后的生理机制的基本问题。我们使用的技术包括：1)在飞行过程中跟踪身体和翅膀运动的高速视频，2)驱动翅膀的肌肉的电子记录，3)参与视觉处理的单个或多个神经元的单通道和多通道神经生理记录，以及4)虚拟现实环境来模拟真实的视觉运动。具体来说，我们研究了可达蝗虫神经系统中识别神经元对行为相关视觉线索的反应。我们感兴趣的是如何处理信息，以生产飞行操纵。飞行是一种复杂的行为，需要快速检测和处理复杂的感官信号。这种行为也由相对较少的神经元控制，使其易于进行严格的实验操作。例如，我们研究识别的视觉神经元如何对多个物体或改变运动轨迹的物体做出反应，这些物体可能代表或不代表对动物的危险。我们小组过去的工作表明，单个已识别的神经元能够提取有关复杂物体运动的信息，而其他已识别的神经元可能有助于更高层次的处理，从而驱动自然的飞行行为。我们还发现，飞行肌肉活动的时间可以预测翅膀的运动和动物在三维空间中的方向。利用这些发现，我们开发了一个回避行为模型，这将进一步完善我们的虚拟现实系统，并扩大我们目前与机器人研究人员的合作。这些研究首次展示了这些结果，并推动了构成这一提议基础的现有假设。通过结合多个神经元的多通道记录和虚拟现实技术，我们将进一步了解这种动物在自然环境中遇到的现实复杂场景的呈现过程中的行为和多神经元反应。因此，我们将有助于理解神经系统如何检测重要的感觉线索，以及如何使用这些线索来产生适当的行为反应。结果对于理解行为的神经控制的基本原理非常重要，并将被纳入人工系统的控制策略中。

Most animals live in complex environments that often contain conflicting information about their surroundings. Survival, therefore, depends on the ability of the nervous system to collect and process relevant information, allowing the animal to generate an appropriate behavioural response. To understand the neural control of behaviour, we must first understand how an animal orients in three-dimensional space and how kinematics of locomotory appendages are driven by relevant muscle activity, which ultimately produces forces that move the animal in its environment. To understand central mechanisms of how behaviour is produced and modified, we must study how the nervous system detects environmental cues and drives appropriate locomotory muscles. While single neurons may be suited to detect specific aspects of the environment, the combined activity of groups of neurons is crucial for detecting and deciphering complex sensory cues. Common detection schemes in animals ranging from insects to humans imply that similar fundamental processes are involved in extracting important information from the environment. By identifying mechanisms by which groups of neurons detect relevant cues from complex stimuli, it will also be possible to design better biologically-inspired artificial systems capable of adaptable, self-guided navigation. My laboratory uses insect flight as an ideal model system to address fundamental questions about physiological mechanisms that underly natural behaviour. We use techniques that include combinations of: 1) high speed video to track body and wing movements during flight, 2) electrical recordings of muscles that drive the wings, 3) single and multichannel neurophysiological recordings of single or multiple neurons involved in visual processing and 4) a virtual-reality environment to emulate real visual motion. Specifically, we study responses of identified neurons in the accessible locust nervous system to behaviourally-relevant visual cues. We are interested in how information is processed for production of flight steering manoeuvres. Flight is a complex behaviour that requires rapid detection and processing of complex sensory signals. This behaviour is also controlled by relatively few neurons, making it accessible to rigourous experimental manipulation. For example, we examine how identified visual neurons respond to multiple objects or objects which change motion trajectory, which may or may not represent a danger to the animal. Past work from our group has revealed that a single, identified neuron is able to extract information about complex object motion and that other identified neurons likely contribute to higher level processing that drives natural flight behaviour. We have also found that timing of flight muscle activity predicts wing movement and animal orientation in 3-dimensional space. Using these findings, we have developed a model for avoidance behaviour that will further refine our virtual reality system and expand our current collaborations with robotics researchers. These were the first studies to show these results and drive current hypotheses that form the basis of this proposal. By combining multichannel recordings of multiple neurons and virtual reality techniques we will further understand behavioural and multineuronal responses during presentation of realistic complex scenes that this animal would encounter in its natural environment. Thus, we will contribute to an understanding of how the nervous system detects important sensory cues and how those cues are used to generate an appropriate behavioural response. Results will be important in understanding fundamental principles of the neural control of behaviour and will be incorporated into control strategies for artificial systems.