【作者】 杨贝贝；

【导师】 阮晓钢；

【作者基本信息】 北京工业大学 ， 模式识别与智能系统， 2005， 硕士

【摘要】 自然生命或生物系统的诸多技能是生物个体在发育成长过程中渐进地形成和发展起来的。技能的形成和发展过程是生物系统的认知过程,是自然生命认知能力的体现。理解和模拟自然生命的认知行为,并将这种行为赋予人工生命或人工系统是认知科学、人工智能和机器人学的重要课题。本文对经典条件反射认知模型及其在运动控制系统中的应用进行了研究,取得的主要成果有: 第一:基于尖峰神经元的条件反射模型。本文建立了一种以尖峰神经元为基本元素的经典条件反射模型(Classical Conditioning Model with Spiking Neurons,CCM_sn)。在CCM_sn模型中,各神经元互联形成具有反射弧结构的神经网络。尖峰神经元用时间来计算和交流信息,这使CCM_sn模型能充分表现经典条件反射对时间的依赖性。论文基于有衰减项的Hebb突触修饰规则和蜜蜂脑部认知结构设计了反映“刺激-响应-强化”特征的强化学习算法,使CCM_sn模型具有经典条件反射行为和认知行为。论文应用CCM_sn模型成功地模拟了习得、刺激间隔效应、遗忘、阻止和二阶条件反射等典型现象。第二:基于CCM_sn模型的运动控制技能认知算法。本文研究了基于CCM_sn模型的运动控制系统（Motor Control System,MCS）和运动控制技能认知算法（Motor Skill Cognitive Algorithm,MSCA）。在运动控制技能的认知过程中,首先建立控制对象模型,用于CCM_sn模型的训练和系统仿真试验;然后根据控制对象和控制目标的具体情况确定认知对象各状态的无条件刺激范围和条件刺激范围;最后进行经典条件反射训练。运动控制技能的形成和发展是生命个体学习和认知的过程,条件反射的形成是一个认知的过程,CCM_sn模型的结构和算法具有神经生理学和认知科学依据,因此MSCA算法能使技能的获取表现出生物系统的发育和进化的自组织特征,使机器主体（Robotic Agent）自发地学会控制技能。第三:基于CCM_sn模型和MSCA算法的倒立摆控制本文以倒立摆为具体控制对象,设计了基于CCM_sn模型和MSCA算法的控制系统。根据倒立摆的具体情况,将CCM_sn模型的反射机构分为正负两部分。根据控制目标,设计了认知评价函数,用于评价CCM_sn模型的认知效果。在实验初期,CCM_sn模型只对无条件刺激产生响应,即只在摆杆超过设定值时才对更多还原

【Abstract】 During the growing up process, natural life or biological system develops various skills gradually. The formation and development of skills are both cognitive process, which are the embodiments of cognitive ability. To understand and simulate the cognitive behaviors of natural life and endue these to artificial life is an important task for the field of cognitive science, artificial intelligence and robotics. This thesis studied the motor skill cognitive model and its applications, and the achievements can be summarized as follows: 1. Classical Conditioning Model with Spiking Neurons A Classical Conditioning Model with Spiking Neurons (CCM_sn) is proposed. The model comprises a number of spiking neurons connecting to form a neural network with reflex arc structure. Spiking neurons compute and communicate by timing, which makes this model fully exhibiting the dependency of classical conditioning on timing. A reinforcement learning method based on the Hebb rule with a decay constant has been designed characterized with a property of ‘stimulate-response-reinforcement’. Our model can successfully stimulate many typical experiments such as acquire, inter-stimulus effects, extinction, blocking, and secondary conditioning. 2. Motor Skill Cognitive Algorithm Based on CCM_sn Motor Control System （MCS） and Motor Skill Cognitive Algorithm （MSCA） are studied, which are both constructed on the base of CCM_sn. The cognition process of motor skill has three steps. First, model the controlled object to train CCM_sn, Secondly, set the range of unconditioned stimulus and conditioned stimulus for each state variable according to the certain plant and the goal. Finally train the cognitive model by classical conditioning experiments. Conditioning is a cognitive process, the structure and algorithm of CCM_sn is set up on the foundation of neurophysiology and cognitive science. As a result, MSCA can endue the skill acquiring process with self-organized property of growth and development in biological system. 3. Inverted Pendulum Control System Based on CCM_sn and MSCA A control system based on CCM_sn and MSCA is designed, and the inverted pendulum is used to test the system. Considering the property of inverted pendulum, CCM_sn is divided into two parts, one is for positive direction and the other is for negative direction. Critical function for Cognition is designed to value the更多还原