RI: Medium: Dynamical Coordination and Sequencing of Multifunctionality in Animals and Robots

如何为自主机器人创造智能控制，使它们能够灵活和适应性地应对不断变化的环境？在动物世界中，相对简单的动物，如软体无脊椎动物，能够协调它们许多可能的动作，随着条件的变化灵活地转移和排序多种行为，并学习根据经验改变它们的行为。然而，对于机器人来说，这仍然是一个挑战，本项目使用一种新型的控制体系结构来解决这一问题，该体系结构可以产生感觉驱动或循环运动。传统的机器人控制体系结构有三层：高层慎重规划，低层反应控制，以及用于排序和简单决策的中层。为智能行为创造中级控制器尤其具有挑战性，也是自主机器人技术进步的主要障碍。这一问题将使用一种新颖的神经启发控制体系结构-稳定的异宿通道(SHCS)来解决，它可以灵活而稳健地协调多个自由度的多功能，并可以轻松地处理行为层次结构、时间决策和学习。它们的特性还允许它们结合两种传统机器人控制方法的一些最佳方面：有限状态机和中央模式生成器(极限环)控制器。将在易于实验的软体动物中分析基于SHC的动力学结构，这些神经生物学结构的原理将用于在新型超冗余软体机器人平台中实现多功能行为。自适应、灵活控制的软体或超冗余机器人有许多可能的应用，它们能够以不同的方式协调它们的多个自由度来完成多种功能。多功能蠕虫状机器人可以在不同直径和与重力成任何角度的管道中爬行，并在十字路口急转弯。一个中空的超冗余度机器人可以从内部检测自来水总管，而不会中断水流。这样的机器人可以用于石油和天然气管道的检查，以避免成本高昂和对环境造成灾难性的泄漏。可以开发更小的版本用于胃肠道的内窥镜诊断。提出的工作将导致一个单一的控制器框架，可以稳健地协调机器人内的多个耦合驱动机构，并描述动物和机器人中不同行为的顺序。

How can intelligent control be created for autonomous robots that will allow them to respond flexibly and adaptively to changing environments? In the animal world, relatively simple animals such as soft-bodied invertebrates, are capable of coordinating their many possible movements, flexibly shifting and sequencing multiple behaviors as conditions change, and learning to alter their behavior based on experience. For robots, however, this remains a challenge, which is addressed in this project using a novel control architecture that can produce sensory driven or cyclic movements.Traditional control architectures for robotics have three layers: high level deliberative planning, low-level reactive control, and an intermediate level for sequencing and simple decision-making. Creating intermediate level controllers for intelligent behavior is particularly challenging, and a major obstacle to progress in autonomous robotics. This problem will be addressed using a novel neural-inspired control architecture, stable heteroclinic channels (SHCs) that can flexibly and robustly orchestrate multiple degrees of freedom for multifunctionality, and can readily handle behavioral hierarchies, temporal decision-making, and learning. Their properties also allow them to incorporate some of the best aspects of the two traditional approaches to robotic control: finite state machines and central pattern generator (limit cycle) controllers. SHC-based dynamical architectures underlying multifunctionality will be analyzed in a soft-bodied animal that is tractable to experimentation, and principles from these neurobiological architectures will be used to implement multifunctional behavior in a novel hyper-redundant, soft-bodied robot platform.There are many possible applications for adaptive, flexibly-controlled soft-bodied, or hyper-redundant, robots that are able to coordinate their many degrees of freedom in varying ways to accomplish multiple functions. A multifunctional worm-like robot could crawl through pipes of varying diameter and at any angle with respect to gravity and make sharp turns at intersections. A hollow hyper-redundant robot could inspect water mains from the inside, without interrupting water flow. Such a robot could be used for oil and gas pipeline inspections to avoid costly and environmentally disastrous leaks. Smaller versions could be developed for endoscopic diagnosis of the gastrointestinal tract. The proposed work will lead to a single controller framework that can robustly coordinate multiple coupled actuated mechanisms within a robot, and describe the sequencing of distinct behaviors in both animals and robots.