| Decision-making and planning are important components of unmanned vehicles,which play the role of the "brain" of autonomous vehicles,and the results directly affect the safety and comfort of vehicles.At present,there are a large number of mature algorithms in this field,which can be basically divided into search-based,optimization-based and machine learningbased categories.However,these algorithms often operate on deterministic environmental models and fully observable environments.However,there are often many uncertainties in the real environment,and this uncertainty mainly comes from two aspects.On the one hand,the sensing range of the sensor is limited,and there are errors in the sensing algorithm itself.On the other hand,the intentions of other traffic participants in the driving environment are unpredictable.That is to say,there are many uncertainties in reality,and self-driving vehicles operate in partially observable environments.Based on the road network relationship provided by OpenDRIVE,A speed planning algorithm considering partial observable environment and an improved hybrid A* path planning algorithm are proposed in this paper,and a simulation software SimTik is constructed to simulate this uncertainty.The speed planning algorithm we proposed modeled the driving environment of automatic driving as a partially observable Markov decision model(POMDP),which could well consider the uncertainties in the driving environment of urban roads and reasonably plan the driving speed under the perception of being blocked by buildings to ensure safety.Based on the road reference line provided by OpenDRIVE,the proposed improved hybrid A* algorithm can adapt to urban structured roads,and its search speed and planning consistency are greatly improved.In order to shield the interference of various sensor models and expose this uncertainty in the environment to the autonomous driving decision planning module,so as to better test the effect of the algorithm on dealing with this unknown,we built our own simulation software SimTik.The software uses ray tracing to simulate the sensing range of the sensor and has a variety of vehicle models and control algorithms built in.In the future,we will open source the software.At the end of the paper,we give the operation effect of the decision-making planning algorithm in the simulation software.At the same time,we have also carried out several real vehicle tests.High security. |
