黄敏, 路飞, 李晓磊, 田国会, 孟可. 基于IHDR算法和BP神经网络复合框架的机器人服务自主认知和发育系统[J]. 机器人, 2019, 41(5): 609-619. DOI: 10.13973/j.cnki.robot.180650
引用本文: 黄敏, 路飞, 李晓磊, 田国会, 孟可. 基于IHDR算法和BP神经网络复合框架的机器人服务自主认知和发育系统[J]. 机器人, 2019, 41(5): 609-619. DOI: 10.13973/j.cnki.robot.180650
HUANG Min, LU Fei, LI Xiaolei, TIAN Guohui, MENG Ke. Autonomous Cognition and Development System of Robot Service Based on a Composite Framework Combining IHDR Algorithm with BP Neural Network[J]. ROBOT, 2019, 41(5): 609-619. DOI: 10.13973/j.cnki.robot.180650
Citation: HUANG Min, LU Fei, LI Xiaolei, TIAN Guohui, MENG Ke. Autonomous Cognition and Development System of Robot Service Based on a Composite Framework Combining IHDR Algorithm with BP Neural Network[J]. ROBOT, 2019, 41(5): 609-619. DOI: 10.13973/j.cnki.robot.180650

基于IHDR算法和BP神经网络复合框架的机器人服务自主认知和发育系统

Autonomous Cognition and Development System of Robot Service Based on a Composite Framework Combining IHDR Algorithm with BP Neural Network

  • 摘要: 为了解决传统的基于知识或基于学习的机器人服务认知机制的智能性和普适性较差的问题,构建了一个基于IHDR(增量分层判别回归)算法和BP(反向传播)神经网络复合框架的机器人服务任务自主认知和自主发育系统.在家庭服务机器人智能空间中丰富的传感器和物联网技术的支持下,采集大量用于机器人学习和发育的样本数据;在此基础上,针对智能空间样本数据的混合特性,设计改进的IHDR算法,实现对混合型样本数据的聚类更新和响应计算,并将生成的IHDR树作为机器人存储历史经验的“大脑”,使机器人能够利用“大脑”中已有的经验进行自主学习和相应判断,以实现对服务的自主认知;利用JSHOP2(Java simple hierarchical planner)规划器对认知的复杂任务进行分解,得到可被机器人直接执行的原子任务.为了避免IHDR树规模不足的局限性,设计基于BP神经网络的服务认知算法,利用样本数据训练BP神经网络,实现智能空间实际场景到用户所需服务的映射,在IHDR树无法提供历史经验的情况下,使机器人仍能基于BP神经网络自主进行服务决策.然后将此映射结果以增量的方式更新到IHDR树中,丰富其具备的经验知识,实现机器人服务自主认知能力的发育.仿真实验结果表明,该复合框架可以有效提高服务机器人对智能空间情景下用户所需服务的认知准确性及认知发育能力,推进人机共融的实现.

     

    Abstract: In order to solve the poor intelligence and universality problems of the home service robot with traditional knowledge-based or learning-based service cognitive mechanisms, an autonomous cognition and development system of robot service tasks based on a composite framework combining incremental hierarchical discriminant regression (IHDR) algorithm with BP (backpropagation) neural network is constructed. A large amount of sample data are collected for learning and development of robots based on the technical support provided by multiple sensors in intelligent space and Internet of Things (IoT). On this basis, a modified IHDR algorithm is designed in light of the mixing characteristics of these sample data to achieve cluster updating and response calculation for mixed-type sample data, and an IHDR tree is constructed as the "brain" of robot to store its historical experience, which will provide historical experience for robot to learn and judge, realizing autonomous cognition of services. The JSHOP2 (Java simple hierarchical planner) is used to decompose the cognized complex tasks to obtain atomic tasks which can be directly executed by the robot. Meanwhile, a service cognition algorithm based on BP neural network is developed to avoid the limitation of IHDR tree size. The BP neural network is trained with sample data to map the actual scene in intelligent space to the service required by user, and thus the robot can still make service decisions autonomously based on BP neural network in case that the IHDR tree can't provide historical experience. Next, the IHDR tree is updated incrementally with the mapping result, enriching the robot's experience and knowledge, and realizing the development of autonomous cognitive ability for robot service. The simulation results show that the accuracy and developmental ability of cognition for services required by the user are improved for the service robot in intelligent space by the composite framework, which may promote the realization of man-machine communion.

     

/

返回文章
返回