运动技能认知模型与认知机器人"感知－行动"环的形成研究

It is the common theme of cognitive science, artificial intelligence, and robotics to explore cognitive mechanisms and comprehend cognitive behaviors, in which artificial systems, especially physical embodied cognitive models including cognitive robots, are playing a more and more important role. To a large extent, the motor skills of humans and animals are acquired. Motor skills refer to two aspects, one is motor learning, the other is motor control. This project aims at developing computing models simulating motor cognitive mechanisms of humans and animals, and endow artificial systems or robots with analogous motor cognitive mechanisms to humans and animals so that they are able to shape and develop perception-action cycles autonomously like humans and animals by motor learning, and be gradually in control of motor skills. We will investigate three sorts of the mechanisms of motor cognition in organisms: (1) shared motor representation, (2) motor priming, and (3) motor learning including imitation induced by motor priming, in order to develop computing models simulating motor cognitive mechanisms. Meanwhile, we will build cognitive robots with bionic receptors and effectors so that they possess perception-action behaviors. This project will apply the cognitive models to the robots so that they possess the behaviors or capability of acquiring motor skills. We will primarily examine three key issues: (i) how to design the artificial sensorimotor systems or perception-action cycles, and make it self-organizing, (ii) how to represent perceptions and code actions in the motor cognitive robots and model the mechanism of shared motor representation, and (iii) how to model the mechanism of motor priming and induce the mechanism of imitation for the motor cognitive robots. This project has scientific significance as well as engineering value. The scientific significance lies in the fact that it helps us to understand the cognitive mechanisms and behaviors to simulate animals with physical or embodied cognitive robots. The engineering value lies in the fact that it helps us to design and make more autonomous, more intelligent, and more adaptive robotic systems, endow them with cognitive capability, and make them serve us better. This project will advance our understanding of the mechanisms of motor cognition of organisms by the embodied experiments on the motor cognitive robots. Its outcome is not only benefit to cognitive science, but to artificial intelligence and robotics as well.

探索认知机理，理解认知行为，并将认知赋予人工系统，是认知科学、人工智能、机器人学等诸多学科共同关心的课题。人和动物的运动技能，在很大程度上是后天习得的，其中包含两方面，一是运动神经学习，二是运动神经控制。课题拟建立模拟人和动物运动技能习得的计算模型，赋予机器人或其它人工系统类似人和动物的运动神经认知机能，使其像人和动物一样，通过运动神经学习，自主地或自组织地形成其"感知－运动"环，渐近地掌握运动控制技能。课题将主要通过模拟运动神经认知中的共享运动神经表达机制、运动神经触发机制及其诱导的模仿行为，建立模拟运动神经认知机制的计算模型，同时，研究和设计具有"感知－行动"能力的认知机器人系统，将运动神经认知模型应用于认知机器人，使其具有类似人和动物的运动技能习得行为。课题将有助于我们理解人和动物运动技能的习得，并将这种认知机制应用于机器人等人工系统，既有益于认知科学，也有益于人工智能和机器人学。

探索认知机理，理解认知行为，并将认知赋予人工系统，是认知科学、人工智能、机器人学等诸多学科共同关心的课题。人和动物的运动技能，在很大程度上是后天习得的，其中包含两方面，一是运动神经学习，二是运动神经控制。课题建立了模拟人和动物运动技能习得的计算模型，赋予机器人或其它人工系统类似人和动物的运动神经认知机能，使其像人和动物一样，通过运动神经学习，自主地或自组织地形成其“感知－运动”环，渐近地掌握运动控制技能。课题研究工作涉及：1）自组织的人工“感知-行动”系统，2）运动神经技能学习机制，3）关于认知发育自动机与行为学习模型的研究，4）具身认知模型（运动技能认知机器人）研究，5）实验研究。在有关自组织的人工“感知-行动”系统的研究方面，课题建立了基于内发动机机制的感知行动认知模型及具有发育机制的感知行动认知模型。在有关运动神经技能学习机制的研究方面，课题分别基于Boltzman机及操作条件反射原理建立认知模型，以蟑螂为仿生对象，再现了昆虫负趋光的运动神经技能学习行为。在有关认知发育自动机的研究方面，课题通过热力学过程来模仿动物的行为学习过程, 并且从蒙特卡罗方法以及 Metropolis 算法, 和模拟退火算法中提出了操作条件反射自动机。在具身认知模型研究及实验研究方面，课题设计构造了物理实验模型，包括仿昆虫的六足机器人及电磁独轮机器人等。课题完成了预定的研究任务，研究工作取得了重要进展和有参考价值的研究成果。课题研究工作将有助于我们理解人和动物运动技能的习得，并将这种认知机制应用于机器人等人工系统中，既有益于认知科学，也有益于人工智能和机器人学及机器人技术。