映射领域可自主收缩的操作条件反射自动机

Operant conditioning automaton with mapping fields of spontaneous contraction

为避免操作条件反射学习模型中存在发生小概率操作行为所引发的不良操作后果,设计了一种映射领域可在线自主收缩移动的操作条件反射自动机,映射领域指机器人各状态映射的操作行为集合,其收缩是通过定义学习误差的界限值和操作行为选取的概率阈值两个指标来实现的。对映射领域可自主收缩的操作条件反射学习的收敛性进行了分析,从理论上证明:操作行为熵随映射领域的收缩收敛至极小。针对两轮机器人运动平衡控制的仿真结果表明,映射领域的收缩性使操作条件反射自动机可以在线地在最有意义的映射领域内搜索学习,通过有效的动态的消除无用的映射区域,提高了系统的学习速度和精度。

The phenomenon of small probabilities operant action exists in operant conditioning learning model and the occurring of small probabilities operant action will lead to bad consequence.To avoid the phenomenon,an operant conditioning automaton is designed in which the mapping fields could contract spontaneously.The mapping field was defined as the mapping operant action set of each state and the contracted mapping field was implemented by defining two index values,i.e.,the bound value of learning error and the threshold value of probability.Theory analysis was made for the convergence of operant conditioning learning of operant conditioning with mapping fields of spontaneous contraction,which theoretically proves that the operant action entropy can converge to minimum with the contraction of mapping fields.The result of the simulation experiment in the motion balanced control of two-wheeled robot show that contracting the mapping action fields in this way automatically eliminates unnecessary operant actions and the automaton can learn in the most significant action fields,thereby increasing the learning speed and learning precision.