Principles of intelligent sensorimotor behavior under informational constraints

大多数人在考虑动物的智力时都会想到学习、解决问题、使用工具、导航和交流。然而，这种对智力的看法正在发生转变。研究人员越来越多地认识到，仅仅与环境互动就需要动物解决非常复杂的问题。事实证明，这些问题中的许多都比经典定义的智能行为更难解决。例如，考虑一只昆虫在不平坦、不可预测的地形上行走。协调六条腿的运动是一个具有挑战性的问题。与环境互动是困难的，因为动物对自己的身体或地形没有完全的了解。它也没有太多的时间来收集信息。当动物与世界互动时，这种信息的缺乏是很常见的。目前，我们还不清楚动物如何应对这种缺乏知识的问题。蝙蝠是研究这个问题的理想动物。他们的声纳系统只能为他们提供关于周围环境的有限信息。尽管如此，它们表现出惊人的协调和聪明的行为。目前的项目调查蝙蝠用回声定位来克服缺乏感觉信息的策略。其目的是使用机器人模拟几种蝙蝠在感觉信息缺乏最严重的条件下的行为，例如，在蝙蝠从洞穴中出现的过程中。使用机器人而不是计算机模拟可以忠实地模仿蝙蝠接收到的声纳信号。通过研究蝙蝠可以用来处理信息不足的策略，该项目旨在更普遍地了解动物是如何克服这个问题的。动物的智力通常以学习、解决问题、使用工具和其他高级认知方面的能力为代表。然而，最近关于动物智力的观点强调，在与物理环境的相互作用中，对行为进行调整以适应时不时出现的意外情况的能力。应对动态的、不可预测的世界是具有挑战性的。最重要的是，因为面对严重的信息不足，行动和感知必须继续进行。在表演时，动物并不完全了解这个世界或他们的身体。该项目提出并测试了通过建立蝙蝠声纳行为的生物机器人模型来处理信息约束的战略，从而能够准确地模拟声纳信号。特别是，有人提出，将特定任务的感觉输入直接映射到动作是处理信息限制的基本原则。回声定位蝙蝠是在信息限制下理解动物智力的理想选择，因为它们的声纳系统只能提供有限的环境信息。然而，回声定位蝙蝠在与复杂环境的互动中表现出惊人的智能行为。理解智能感觉运动控制的原理可能有助于理解动物的所谓高级认知功能。这些与感觉运动行为并不是不连续的。事实上，要理解动物智能(以及潜在的人类智能)的基本原理，可能需要了解它们的感觉运动控制起源。因此，研究支持脊椎动物感觉运动智能的原理可能有助于解释与智能有关的更经典的功能(例如，工具使用和规划)。因此，这项建议旨在帮助更深入地了解动物的智力和认知。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Most people think about learning, problem-solving, tool-use, navigation, and communication when considering animal intelligence. However, this view of intelligence is shifting. Increasingly, researchers recognize that merely interacting with the environment requires animals to solve very complex problems. Many of these problems turn out to be more challenging to solve than what is classically defined as intelligent behavior. For example, consider an insect walking across uneven, unpredictable terrain. Coordinating the movement of six legs is a challenging problem. Interacting with the environment is difficult because the animal does not have complete knowledge about its body or the terrain. Nor does it have much time to gather information. This lack of information is common when animals interact with the world. Currently, we do not have a good idea of how animals deal with this lack of knowledge. Bats are ideal animals in which to study this problem. Their sonar system provides them only with limited information about their surroundings. Despite this, they perform astoundingly coordinated and intelligent behavior. The current project investigates strategies echolocating bats use to overcome the lack of sensory information. The aim is to use robots to model several bat behaviors under conditions where the lack of sensory information is the most severe, for example, during the emergence from caves. Using robots instead of computer simulations allows mimicking the sonar signals received by bats faithfully. By examining strategies bats can use to deal with informational deficiencies, the project aims at understanding how animals more generally can overcome this problem.Animal intelligence is often typified by abilities such as learning, problem-solving, tool use, and other facets of higher-order cognition. However, more recent views on animal intelligence emphasize the ability to adjust behavior to the moment-to-moment contingencies arising in interaction with the physical environment. Coping with the dynamic, unpredictable world is challenging. Most importantly, because action and perception have to proceed in the face of substantial informational deficiency. While acting, animals do not have complete knowledge about the world or their bodies. This project proposes and tests strategies for dealing with informational constraints by building bio-robotic models of bat sonar behavior, allowing for veracious mimicking of sonar signals. In particular, it is proposed that direct mapping of task-specific sensory input to action is a fundamental principle for dealing with informational constraints. Echolocating bats are ideal for understanding animal intelligence under informational constraints because their sonar system only provides limited information about the environment. Nevertheless, echolocating bats perform astounding intelligent behavior in interaction with complex environments. Understanding principles for intelligent sensorimotor control might help to understand the so-called higher cognitive functions in animals. These are not discontinuous with sensorimotor behavior. Indeed, understanding the fundamental principles of animal intelligence (and potentially human intelligence) may require learning about their sensorimotor control origins. As a result, investigating the principles supporting vertebrate sensorimotor intelligence might help explain functions more classically associated with intelligence (e.g., tool use and planning). As such, this proposal aims at contributing to a deeper understanding of animal intelligence and cognition.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.