视觉陌生度驱动的增量自主式视觉学习算法

针对传统机器学习框架下设计智能机器人造成的视觉任务执行时学习主动性差、对不确定情况适应性差、知识与能力扩展性差等问题,立足近年来新提出的认知发育思想,提出一种由视觉陌生度驱动的增量自主式视觉学习算法。算法根据在线主成分分析(PCA)计算视觉陌生度,作为Q学习内部动机,以PCA子空间的更新作为知识的主动学习与积累,并由以视觉陌生度为内部动机的Q学习引导,使得机器人能根据所学知识与所"见"场景的陌生程度来决策下一步如何学习。实验结果表明,该算法具有自主探索与学习性能、主动引导机器人学习新知识的能力,以及在线、增量地获取积累知识并发育其智能的能力。

Visual novelty driven incremental and autonomous visual learning algorithm

In intelligent robot design,the traditional machine learning paradigm is commonly used.However,the traditional methods cause problems in visual tasks such as low learning initiative,lack of adaptability with uncertainty and bad expansibility of knowledge and ability.According to the new research direction called cognitive development learning,a visual novelty driven incremental and autonomous visual learning algorithm is proposed,in which the internal motivation is defined as visual novelty which is calculated by online PCA.The autonomous learning and accumulation of knowledge is implemented in the form of updating PCA subspace,which is guided by internally motivated Q-learning using visual novelty.Equipped with the proposed algorithm,a robot makes the next learning decision by judging the novelty between learned knowledge and what is seen now.Experimental results show that the algorithm has the ability of autonomous exploring and learning,actively guiding the robot to learn new knowledge,acquire knowledge and develop intelligence online and in incremental manner.