Research On Four-rotor Controller-in-the-loop Simulation System For Warehouse Inspection Based On Unity3D

The quadcopter has become a research hotspot for warehouse inspection due to its low price,compact size,flexible maneuverability,and outstanding reliability.However,due to the complexity of warehouse scenarios,UAV is not conducive to actual flight during development,making it difficult to obtain reliable experimental data.Therefore,in response to the challenges of development difficulties,long cycles,and high costs for quadcopters applied in warehouse inspection,this project was developed a quadcopter control simulator based on Unity3D for warehouse inspection,providing an algorithm verification platform that does not require actual flight.The main research focus includes real-time rendering of three-dimensional scenes based on Unity3D,optimization of the realism of virtual scene visual images,and quadcopter control simulation technology,which can support the verification of the bottom-level control algorithms and upper-level image application algorithms of the quadcopter controller.(1)Based on the 3D modeling software and referring to a certain elevated warehouse,a 1:1virtual test scene was established by using polygon modeling technology.However,due to the complexity of the virtual scene and the large number of geometric objects,the real-time operation of the system is severely affected.To address this,an optimization method for the three-dimensional modeling software model was studied.The method was employed a model geometry reduction approach to reduce the number of triangles in the model and eliminate hidden vertices without altering the appearance of the model,thus optimizing the scene data.Additionally,a static batch processing based on CPU was studied,which consolidates models with the same material in a new mesh,reducing the number of Draw Calls,lowering memory consumption,and improving the real-time performance of the CPU.Finally,GPU-based occlusion culling is also studied,which uses a virtual camera to obtain the depth data Z-Buffer of the model,removes occluded models,reduces VRAM consumption,and significantly improves the real-time performance of the GPU.This real-time optimization method ensures that the scene runs at 60 frames per second or above,meeting the real-time requirements.(2)Due to significant discrepancies between images captured by virtual vision sensors and those captured by real vision sensors,experiments related to image processing can be adversely affected.This thesis addresses this issue by studying a physically-based rendering approach.It was employed a BRDF lighting model based on physical principles,with precise parameters for specular and diffuse reflection that accurately reflect the real world.The use of normal texture mapping in tangent space restores the model’s true details.Delayed rendering,in conjunction with the URP rendering pipeline and SSAO technology,enhances the contrast of the images displayed on the screen.HDR technology maps high-range luminance to low-range luminance,capturing the high contrast of the real world and comprehensively improving the visual realism.Finally,noise is added to the camera to simulate the actual process of image acquisition.(3)To address the issue of unreliable experimental data resulting from inaccurate simulation modules,this thesis proposes a precise hardware-in-the-loop simulation framework that combines the design principles of the physical world.A realistic training environment is built in Unity3D with support from Airsim,incorporating quadcopter modules,environment modules,physics engines,sensor modules,rendering modules,and API layers.Through this simulation platform,the control part of the quadcopter hardware can be calculated in real-time and the calculation results are transmitted to the drone module via Pixhawk.The drone module executes flight tasks based on the provided realistic drone training scene and feeds back the execution results to the controller,enabling closed-loop experiments that provide robust support for the development of underlying control algorithms.(4)In order to further validate the key technologies of this study,a simulation of the inventory-taking process using a quadrotor in an elevated warehouse scene was conducted.The pose information of the quadrotor during its flight was recorded to validate the reliability of the underlying control algorithm.The images captured by the quadrotor were used for cargo quantity detection to verify the authenticity and effectiveness of the virtual images,as well as to test the feasibility of the upper-level application algorithm.