摘要
近年来,无人驾驶无论Tesla路线、Waymo路线,还是Intel路线等,都重点关注感知、规划和决策等计算认知,而交互认知严重缺失.这一问题不解决,自驾驶车难以被社会接受与认可.本文分析了自驾驶车交互认知的丰富性和实现难度,提出基于自然语言交互的车载智能音箱对话、肢体语言交互和车体语言交互3种交互形式.通过智能音箱及互联网实现了自驾驶车与车主、乘员、运维人员、开发人员、远程服务请求等之间的交互;解决了自驾驶车对路边行人和执勤交警的手势识别、理解与应对;攻克了狭路会车和超车并道两种常见车体交互的难题.最后,基于丰富的自驾驶交互认知形式,给出独立于决策总线的交互总线架构设计,并在多类型智能车上开展应用.
In recent years,self-driving approaches,such as Tesla,Waymo,and Intel,have focused on computational awareness,including cognitive perception,planning,and decision making.However,little effort has been concentrated on interactive cognition.This problem is important for unmanned vehicles to be accepted by society.Targeting on this urgent issue,we analyze the richness and difficulties of the interactive cognition of self-driving vehicles.The interactive cognition of self-driving can be divided into the vehicle intelligent speaker dialogue based on natural language interaction,body language interaction,and vehicle body language interaction.Through the interactive cognition of self-driving,it can employ intelligent speakers and the Internet to achieve the interaction between self-driving vehicles and owners,crews,operation and maintenance personnel,developers,remote service requests,etc.This paper solves the gesture recognition and understanding of pedestrians and road traffic police,and also overcomes two typical body interaction problems,including the car-meeting task at narrow roads and the car overtaking/merging task.Finally,based on many self-driving interactive cognitions,we provide an interactive bus architecture that is independent of a decision bus,and apply it in various types of smart cars.
作者
马楠
高跃
李佳洪
李德毅
Nan MA;Yue GAO;Jiahong LI;Deyi LI(School of Robotics,Beijing Union University,Beijing 100101,China;School of Software,Tsinghua University,Beijing 100084,China;Academy of Military Science,Beijing 100036,China)
出处
《中国科学：信息科学》
CSCD
北大核心
2018年第8期1083-1096,共14页
Scientia Sinica(Informationis)
基金
国家自然科学基金(批准号:61671267
61672178)
英国皇家工程院牛顿基金项目(批准号:UK-CIAPP\324)
北京市自然科学基金(批准号:4182022)
北京联合大学2017年度人才强校百杰计划(批准号:BPHR2017CZ10)资助项目
关键词
拟人驾驶
多总线架构设计
车载智能音箱
手势识别与理解
车体交互
personnel driving
multi-bus architecture design
intelligent car speaker
gesture recognition and understanding
body interaction
