基于非合作博弈威胁度的智能驾驶策略研究

Research on Intelligence Driving Strategy based on Competition Threaten Risk Degree of Non-Cooperative Game Theory

本文基于非合作博弈竞争威胁风险度(CTRD),提出了在风险驾驶场景中的智能驾驶决策及执行算法系统。基于实时采集的博弈双方直接和间接信息,进行浅层和深层分析,计算每次竞争行为的CTRD,并在持续高频的连续博弈场景中对博弈威胁风险水平(CTRL)综合度量分类,以此建立驾驶员行为风险评估模型并标签化处理。此方法可描述具备较强相关性和行为逻辑的交互竞争,通过理性妥协以避免攻击性鲁莽驾驶行为造成的碰撞事故,在保障系统对危险驾驶场景及行为做出安全决策基础上提高通行效率。实验采用3 000多组实际道路的切入数据,对其中涉及博弈竞争行为进行比较分析。结果显示,在连续交互非合作博弈回合中,面对竞争威胁挑战,基于CTRD的驾驶策略预测与驾驶员实际行动呈相关性分布,标签化的CTRL对实际博弈行为预测符合度为94%。

An intelligent driving strategy system for critical scenario based on competition threaten risk degree(CTRD) of noncooperative game theory(NCGT) is proposed.By real-time collection of direct/indirect information with shallow and deep analysis,the CTRD of competing actions are calculated.In its ongoing potentially critical scenario of high occurrence rate,competition threat risk level(CTRL) is measured to establish a driver behavior risk evaluation model with labeling.The system can describe the interactive competition with strong correlation and behavioral logic,and could avoid clash accident caused by aggressive,reckless driving behavior with rational compromise.More than 3 000 groups real-road cut-in data is employed to evaluate the feedback competition after facing threat challenges,the result shows the labeled CTRL prediction for the actual behavior is 94% in compliance.It presents the correlation distribution between driving strategy prediction and actual operations of the tentative and feedback in the continuous non-cooperative competition rounds.