Perception and Control: A Dynamic Perspective

智能系统需要通过明智地选择稀缺资源来控制信息流。在生物学和工程学中，运动和形状估计可能是视觉通路需要执行的最重要的子任务。在这项研究中，我们专注于运动，研究人造海龟如何在视觉上照顾它的猎物，或者在工程上，机器人如何控制它的凝视并锁定移动的目标。我们开始的问题是如何生成合理的运动模型来描述视网膜上的光流，视网膜流如何由神经信号表示，以及如何使用对一系列噪声观察到的尖峰信号进行操作的鲁棒滤波来估计这些运动参数。我们的第二个问题是研究光流如何被视觉通路内化。这包括通过神经电路表示模拟信号，稳定地传递神经代码，从视网膜到皮质，最后在每个阶段进行计算，但主要是在皮质。我们的第三个问题是动态观察大脑皮层，研究不同细胞之间的反馈和前馈相互联系。我们介绍了一种新的具有自动增益控制的神经动力学系统，并计划将我们的模型与已知的实验数据进行比较。我们想知道在什么条件下大脑皮层仍然是一个稳定的动力系统。我们还建议研究大脑皮层互连的频率响应。我们的最后一个问题是研究与眼睛运动有关的运动控制问题。如果海龟必须盯着它的猎物，猎物的瞬时位置和眼动控制系统的输入都是模拟信号，必须用神经代码来表示。这种跟踪信号和控制信号都被神经表示的跟踪问题从未被研究过。具体地说，我们希望设计一种神经电路，它可以产生所需的控制输入，以保持给定的眼睛位置或凝视给定的目标。本研究是关于动力学和控制的，通过神经编码、神经元间通信和计算来实现，最终目标是在动态环境中智能地感知和行动

An intelligent system needs to control the flow of information through judicious choice of its scarce resources. In Biology and in Engineering, Motion and Shape Estimation is probably the most important subtask that the visual pathway needs to perform. In this research, we concentrate on MOTION and study how an Artificial Turtle might visually attend to its prey or in engineering, how a robot might control its gaze and lock on to a moving target. Our starting problem is the engineering problem of generating reasonable motion models that describe the optical flow on the retina, how the retinal flow is represented by neural signals and how these motion parameters can be estimated using robust filters operating on a sequence of noisy observed spikes. Our second problem, is to study how the optical flow is internalized by the visual pathway. This involves representation of an analog signal by the neural circuit, stable communication of the neural code, from retina to cortex and finally computation at each stage but predominantly at the cortex. Our third problem is to view the cortex dynamically and study the feedback and feedforward interconnection of the various cells. We introduce a new neurodynamical system with Automatic Gain Control and plan to compare our model with known experimental data. We would like to know under what condition the cortex remains a stable dynamical system. We also propose to study the frequency response of the cortical interconnection. Our last problem is to study the motor control problem pertaining to the eye movement. If the turtle has to stare at its prey, both the instantaneous position of the prey and in input to the eye-movement control system are analog signals that have to be represented by neural codes. Such a tracking problem with both the tracking signals and the control signals are represented neurally has never been studied. Specifically, we would like to design a neural circuit that would generate the required control inputs to maintain a given eye position or to gaze at a given target.This research is about Dynamics and Control with neural encoding, interneuron communication and computation with the eventual goal to perceive and act intelligently in a dynamic environment