High Performance Computing Simulation Platform For Intelligent Connected Vehicles

Artificial intelligence technology has great application potential in solving the problem of perception and decision-making of intelligent networked vehicles in complex scenes,but at the same time,it faces three challenges,which are the limitation of vehicle computing power,the low development efficiency and the difficulty of scene verification.Thus,a simulation platform based on high performance computing platform is proposed in this paper.The platform mainly includes the following two parts:(1)build a simulation platform based on high-performance computing,and integrate a variety of commonly used artificial intelligence algorithms in it to provide algorithm support;(2)propose a virtual real fusion semi real simulation test scenario,to solve the randomness and authenticity requirements of the scenario in the process of algorithm verification,to improve the security of the test while improving the security of the test Test efficiency.The main research contents of this paper are as follows:(1)Focusing on the problems of high threshold and low efficiency in the development of artificial intelligence algorithm,as well as the problem that artificial intelligence algorithm needs high vehicle computing power,this paper builds a high-performance computing simulation platform integrating low delay image transmission algorithm and a variety of artificial intelligence algorithms.For reducing the threshold and improving the efficiency of AI algorithm development,an AI algorithm library is developed.At the same time,the algorithm library is multi-threaded to support the cloud platform multi vehicle awareness decision-making needs.The algorithm base mainly includes target recognition algorithm,semantic segmentation algorithm,lane line recognition algorithm,end-to-end decision algorithm,finite state machine decision algorithm and low delay image transmission system.Relying on high bandwidth and low delay communication means such as 5,the tasks requiring more computing power such as sensing decision-making in the cloud are realized,and the results are returned to the vehicle end.In this architecture,only the sensor system of the sensing system and the wire control execution system are needed to get off the train in an ideal state,which not only reduces the demand for computing power at the end of the train,but also realizes the ”car road” collaborative sensing in the cloud,so as to improve the driving safety.The basic architecture of the platform mainly includes:high performance computing server,parallel computing acceleration driver,development environment management system based on container engine and encapsulated algorithm library.(2)In order to solve the problem that the artificial intelligence algorithm is difficult to verify under the condition of real vehicle and road,this paper proposes a dynamic traffic flow generation method in the simulation scene.Based on the concept of ”parallel test”and the completed simulation platform and static simulation scenario,this paper constructs a virtual real fusion semi real simulation test scenario,synchronizes the real traffic flow data into the simulation environment,solves the requirements of the scene's randomicity and authenticity in the process of algorithm verification,improves the test safety and improves the test efficiency.In the construction of the semi real simulation scene system of vehicle road cooperation,it is faced with the problem of understanding the road target scene.For this reason,the paper combines two key technologies of ”end sensor” and”edge computing” to dynamically generate traffic flow scene information close to the real scene in the virtual static road scene.(3)By the parallel test of the simulation platform and the real vehicle,the main functions and key performance of the system proposed in this paper are verified.Through the synchronous test of multi-agent in the semi real simulation environment,the platform algorithm is verified to be high-performance computing power and efficiency of the algorithm test.Through the multi-channel real video data return and centralized calculation in the high-performance computing platform,the feasibility and performance of cloud sensing computing system based on super-resolution algorithm and algorithm library are verified.By identifying and locating the target on the real road,and finally generating it dynamically in the virtual scene,the effectiveness of generating the semi real simulation test scene is verified.