认知推理的家庭服务机器人演示学习研究

Research in Learning from Demonstration Based on Cognitive Reasoning for Family Service Robot

提出一种基于对比示范的演示学习方法,旨在指导家庭服务机器人快速智能地学习指定任务决策.在模拟家庭环境下,首先利用视觉手段检测和跟踪任务对象,获取"教师"演示正反例关键信息,通过动作识别和状态检测将信息抽象成符号形式,然后经过认知推理得到演示任务的执行决策和约束条件,并通过动作共享指导机器人重现任务.给定演示实例,在真实移动机器人平台上进行了多次成功演示验证,结果表明本方法在家庭服务机器人快速准确的任务学习方面是有效的.

A new method of Learning from Demonstration based on comparative demons is proposed in this paper, in order to make the robot＇s task learning procedure faster in a simulated household environment. Given an instruction of task learning, the robot de- tects and tracks the targets by computer vision in each instance that the teacher demonstrates, and then converts key information of the task into symbolic presentations through action recognition and state exploration. At last, derives task-related decisions and constraints using cognitive reasoning and transforms them into predications of instructions to guide the robot＇s behaviors. With the new knowl- edge learnt, the robot successfully fulfilled the same given task under several different simulated settings in a reproducible way. It shows that the mentioned means is effective in the learning procedure under home environment.