Research And Implementation Of Key Technologies Of Parallel Simulation Of Swarm UAVs

As the scope of the battlefield and the difficulty of the military operation increases,using only a single UAV to perform tasks will bring about some problems.It may limited by the observation angle and can not achieve full-range detection and attack.A single UAV's failure will also result in the suspension of the entire mission.As a consequence of these problems,utilization of swarm robotics to carry out the tasks is an inevitable trend of future operations.In recent years,the simulation technology of swarm robotics experienced rapid development.However,due to the limited hardware resources of a single compute node and the computational complexity of swarm-robotic simulation,the simulation speed drops significantly as the swarm size increases.It is of great theoretical and practical significance to study how to parallelize the current simulation platform and utilize the hardware resources of multi-node cluster to realize the real time simulation of large-scale swarm robotics.The Gazebo platform's simulation algorithm and code architecture are fully understood through code analysis.And bottleneck analysis are carried out on two typical scenarios(swarm robotics scenary and large campus senary)with Intel Vtune Amplifier.The analysis of bottleneck shows that the simulation performance of single compute nodes is mainly caused by the CPU resource competition between threads.So,multi-node parallelization is a feasible way to improve simulation performance.Aiming at solving the problems that the swarm-robotic parallel simulation will lead to,this paper designs and realizes the solutions based on inter-process synchronization,unique symbol and setting overlap area.This paper also designs and implements a communication algorithm that can transfer variable length messages containing different data types and different models between nodes.This communication algorithm is the basis of parallel simulation across nodes.Based on the cross-node communication algorithm,this paper designs and implements two parallel simulation algorithms suitable for different tasks: a cross-node parallel simulation based on region division and a cross-node parallel simulation algorithm based on UAV's number.Finally,this paper build a multi-node parallel simulation test environment.The quadrotor model is used as the component of swarm robotics.Performance testing is conducted on this test environment through multiple scenarios.The experimental results show that the cross-node parallel simulation algorithm based on region division can effectively improve the simulation performance of the swarm-robotic simulation.The number-based cross-node parallel simulation algorithm needs to further optimize the collision detection part to improve the simulation efficiency.