关于我 | Phillweston Blog

type

status

date

slug

summary

中文版本

2022年，我毕业于西安电子科技大学电子工程专业。

西安电子科技大学是一个在电子信息与通信和计算机领域具有国内较强实力的院校。我在学校期间学习了众多以电子与信息为导向的基础和专业课程，如：高等数学、线性代数、概率论、C语言程序设计、数据结构、数字电路、模拟电路、微机原理、信号与系统、自动控制原理、数字图像处理、信号完整性分析、电磁场理论、微波、通信原理、FPGA逻辑电路设计等。在专业学习的过程中，我积极参加专业课的实验内容，并通过自己的争取和导师的推荐最终加入了E实验楼电子电工实验室和空间物理研究实验室，并开始了我的实验和课题项目的征程。

2018年我初入大学校园的时候，内心还怀揣着对于未来飞行器设计的梦想，加入了RoboMaster机器人战队，以及西电航空协会社团，将自己初步的竞赛方向拟定为机器视觉与人工智能。2019年年初，我提出了“乾天御爪——无人值守化无人机线路巡检系统”，将无人机的天空之眼与起降平台的坚实的后盾融合，创新地将5G万物互联应用与多无人机空中组网和与起降平台通信，使用华为云Hilens端云协同功能完成线路巡检系统的云端训练和本地部署，并最终通过起降平台的扩频通信链路传输至无人机中，同时无人机使用realsense获取下视深度图像与机载IMU数据融合完成精准地形匹配与降落。后面随着项目的发展，我提出将产品应用于电力，河道巡检等领域，并参加了2019年的互联网+创新创业大赛和北斗杯国赛以及相关国创等方向，但由于当时项目继续发展资金受限，以及中期持续投入成本巨大，项目最后仅发展为一代雏形，并最后代表西电参加创新创业展出。事实证明，2022年下半年，DJI大疆创新推广了无人值守化机场以及配套的经纬M30T系列机型，方向与我们团队之前涉及的方向较为类似，该产品推出后大获成功，并最终引领DJI进一步占领工业无人机市场的份额。

2020年，在上一年的乾天御爪项目基础上，我提出了“翼驰协防——空地协同一体化安防系统”，融合了“多无人机全景图动态特征拼接”，“协同化编队控制”等技术方向。为了解决以上技术难题，我查询了很多专业资料，包括视觉SLAM、捷联惯导原理、最优化估计算法、矩阵分析、统计学习方法等。并与孙伟教授ICRA团队硕博学生探讨技术可行性，最终将以上两个方向攻破并发表两篇发明专利，同时参加了2020年的互联网+创新创业大赛和ICRA-RoboMaster人工智能挑战赛（该比赛为DJI与IEEE共同举办的国际类型人工智能挑战赛）。

2020年后，由于疫情的影响，我们专业的很多课程不得不从线下转为线上，因此我也拥有了更多的时间在家或在学校学习课外内的知识。考虑到学校所学的专业理论性较强，我加入上述两个实验室之后通过自己课外时间学习了包括计算机系统与组成原理，C++和Java语言程序设计，SLAM十四讲，机器人中的状态估计，最优化理论，惯性导航系统，现代控制理论，强化学习与策略优化，机器人中的数值优化，计算机视觉中的多视图几何等，并将我的未来发展方向拟定为控制或视觉算法研究。在实验室期间，我利用课后时间参加算法和软件类相关竞赛以此检验我的自学效果，并也因此多次获得国内比赛数次奖项。经过一段时间的竞赛参与，空间院的孙伟教授和电子工程实验室的王新怀教授均对我的自学能力和项目能力表示赞许，并逐渐引导我参与导师们的核心课题，其中包括微波雷达无人机项目以及六旋翼空地协同无人机开发项目，而后我与西安电子科技大学智能院侯教授合作推广了我们2020年互联网+创新创业大赛的空地协同项目和无人机产品，并最终在本校创业创新评估会议中，获得到西电副校长石光明校长的一致赞许。

Ardupilot是全球最大的开源飞控项目，最早由苏黎世理工大学控制团队开源并最后成为开源飞控界的标杆。由于我上一年在“翼驰协防”项目中对于飞行编队控制的探索，我逐渐意识到需要一个替代传统PID的抗扰动算法，后学习现代控制理论中了解到对于扰动模型的观测和补偿，我于是萌生出使用二阶的自抗扰控制（ADRC）算法替代传统的PID算法完成旋翼飞行器的从内环到外环的反馈控制的想法。而后我将此思路用C++代码编写并整合至Ardupilot工程中，并完成了基于ROS Gazebo的仿真测试和基于四旋翼、VTOL 4+1垂直起降飞行器的实际飞行测试，最终我将代码提交commit至Ardupilot其中一个研发分支中。在上述开源项目的开发和学习中，我也进一步巩固了我的软件开发基础，加深了我对于扰动和观测模型的理解，同时学习Ardupilot开源代码也使我进一步了解开源飞控的底层架构和设计逻辑，提高了我对于复杂框架和业务的独立维护能力。

2021年4月份我参与了浙江大学fast-lab的实习，这个实习机会是通过孙老师引荐获得的，实习主要工作是学习VINS-Fusion算法在无人机室内定位的应用，以及学习了ego-planner算法，代替Hybrid A*完成室内无人机自主规划路径的相关任务。实习结束后我将我所学和掌握到的相关内容应用到2021年的工程训练无人机自主飞行赛道中。

2022年，本科即将毕业的我，将毕业课题选定为将视觉与控制融合的方向，与西安一家公司合作完成了校企联合项目，名为“基于无人机的自主航点飞行与双光热成像目标监测算法研究”，融合了我对于前任无人值守项目的理解和对于视觉目标检测算法的研究。在项目的设计和开展中，我使用DJI御系列行业进阶版飞行器，基于DJI Mobile SDK编写了MAVLink航点协议转换，地面特征视觉引导降落（PnP算法），以及基于KCF算法的目标追踪和YoloV5的热成像目标检测与分类。最终在毕业设计答辩中获得较为优秀的荣誉。

2023年年初，我加入了成都纵横自动化企业。成都纵横自动化是全国少有的以垂直起降无人机为主营研发业务的公司，细分领域为：飞控与制导产品，无人机整机和模块业务等。领导对于我的开源飞控系统设计能力和维护能力较为满意，因此将我安排为控制软件和系统开发。在公司工作期间，我参与了数个研究院和国内外的课题项目（包括绵阳空气动力研究院仿真设计，哈工大一站四机编队系统，阿曼国家客户）。尤其是在阿曼国家客户培训中，我凭借自己对于控制系统的理解，出色地完成了为期一个月的全英文客户培训（主要范围：飞控产品架构，机载飞行软件和飞行SDK开发平台，飞控仿真器等），并得到了客户的一致好评。

English Version

In 2022, I graduated from XiDian University with the major subject of electrical engineering.

Having graduated from the Electronic Information and Engineering program at XiDian University, a renowned institution in the fields of electronic information, communications, and computer science, I have studied a wide array of courses focused on electronics and information, such as Advanced Mathematics, Linear Algebra, Probability Theory, C Programming, Data Structures, Digital Circuits, Analog Circuits, Microcomputer Principles, Signals and Systems, Automatic Control Principles, Digital Image Processing, Signal Integrity Analysis, Electromagnetic Field Theory, Microwaves, Communication Principles, FPGA Logic Circuit Design, among others. During my studies, I actively participated in laboratory experiments related to my courses, and through my own efforts and recommendations from my professors, I joined the E-Experimental Building's Electronic and Electrical Engineering Laboratory and the Space Physics Research Laboratory, embarking on my journey of experiments and academic projects.

In 2018, when I first started university, harboring dreams of future aircraft design, I joined the RoboMaster Robot Team and the Xidian University Aviation Association, initially setting my competition focus on machine vision and artificial intelligence. In early 2019, I proposed the "Qian Tian Yu Claw - Unmanned Aerial Vehicle (UAV) Line Inspection System", which innovatively integrated the 'eye in the sky' of drones with a solid backing landing platform. It combined the application of 5G Internet of Things with multi-drone aerial networking and communication with the landing platform. Using Huawei Cloud Hilens for cloud training and local deployment of the line inspection system, it ultimately transmitted data through the landing platform's spread spectrum communication link to the drone, which used realsense to acquire downward depth images and merged them with onboard IMU data for precise terrain matching and landing. As the project progressed, I suggested its application in power and river inspection fields, participating in the 2019 Internet+ Innovation and Entrepreneurship Competition, Beidou Cup National Contest, and other national innovation directions. However, due to limited funding for further development and substantial ongoing investment costs, the project eventually only developed into an initial prototype and was later showcased at Xidian University's Innovation and Entrepreneurship Exhibition. In fact, in the second half of 2022, DJI Innovations launched an unattended airport and the compatible Matrice M30T series models, which were similar to our team's previous direction. This product was a great success after its launch, ultimately leading DJI to further occupy a share of the industrial drone market.

In 2020, building on the previous year's Qian Tian Yu Claw project, I proposed the "Yi Chi Co-Defense - Integrated Air-Ground Security System," incorporating technologies like "Multi-Drone Panoramic Image Dynamic Feature Stitching" and "Coordinated Formation Control." To solve these technical challenges, I consulted extensive professional literature, including Visual SLAM, Inertial Navigation Principles, Optimization Estimation Algorithms, Matrix Analysis, Statistical Studying Methods, and more. I discussed the technical feasibility with Professor Sun Wei's ICRA team graduate students, ultimately overcoming these challenges and publishing two invention patents. I also participated in the 2020 Internet+ Innovation and Entrepreneurship Competition and the ICRA-RoboMaster AI Challenge (an international AI competition co-hosted by DJI and IEEE).

Following 2020, the pandemic led to many of our courses moving online, giving me more time to study extracurricular knowledge at home or at school. Considering the strong theoretical nature of the school's curriculum, after joining the aforementioned laboratories, I spent my extracurricular time learning topics including Computer System and Composition Principles, C++ and Java Programming, 14 Lectures on SLAM, State Estimation in Robotics, Optimization Theory, Inertial Navigation Systems, Modern Control Theory, Reinforcement Learning and Policy Optimization, Numerical Optimization in Robotics, Multi-View Geometry in Computer Vision, and more, setting my future direction in control or vision algorithm research. During my time in the laboratories, I participated in algorithm and software-related competitions to test the effects of my self-study, winning several national awards. After a period of participating in competitions, Professors Sun Wei from the Space Institute and Wang Xinhuai from the Electronic Engineering Laboratory praised my self-learning and project capabilities, gradually guiding me to participate in their core projects, including microwave radar drone projects and six-rotor air-ground cooperative drone development projects. Later, I collaborated with Professor Hou from Xidian University's Intelligence Institute to promote our 2020 Internet+ Innovation and Entrepreneurship Competition air-ground cooperative project and drone products, ultimately receiving unanimous praise from Xidian University Vice President Shi Guangming.

Ardupilot, the world's largest open-source flight control project, was originally open-sourced by the control team of ETH Zurich and has since become a benchmark in the open-source flight control community. Because of my exploration of flight formation control in the previous year's "Yi Chi Co-Defense" project, I gradually realized the need for an anti-disturbance algorithm to replace the traditional PID algorithm. After learning about disturbance model observation and compensation in modern control theory, I decided to use a second-order Active Disturbance Rejection Control (ADRC) algorithm to replace the traditional PID algorithm for feedback control from the inner to the outer loop of rotorcraft. I then coded this idea in C++ and integrated it into the Ardupilot project, completing simulation tests based on ROS Gazebo and real flight tests using quadrotors and VTOL 4+1 vertical takeoff and landing aircraft. Ultimately, I committed the code to one of Ardupilot's development branches. Through my development and learning in this open-source project, I further solidified my software development foundation, deepened my understanding of disturbance and observation models, and learned more about the underlying architecture and design logic of open-source flight control, enhancing my ability to independently maintain complex frameworks and operations.

In April 2021, I participated in an internship at Zhejiang University's fast-lab, an opportunity I obtained through a referral from Professor Sun. The main focus of the internship was learning the application of the VINS-Fusion algorithm in indoor positioning of drones and studying the ego-planner algorithm to replace Hybrid A* for the task of autonomous indoor drone path planning. After the internship, I applied what I learned and mastered to the 2021 Engineering Training UAV Autonomous Flight Track.

In 2022, as an undergraduate nearing graduation, I chose a project that integrated vision and control for my final thesis. This project was a collaborative effort between a company in Xi'an and my university, titled 'Research on Autonomous Waypoint Flight and Dual Optical-Thermal Imaging Target Monitoring Algorithms Based on Drones. It combined my understanding of previous unattended projects with my research on visual target detection algorithms. In the design and execution of the project, I utilized the DJI Mavic series industry advanced version drone, and based on the DJI Mobile SDK, I wrote the MAVLink waypoint protocol conversion, ground feature visual guidance landing (PnP algorithm), as well as target tracking based on the KCF algorithm and thermal imaging target detection and classification using YoloV5. Ultimately, I received a commendable honor in my graduation project defense.

In early 2023, I joined Chengdu Zongheng Automation. Chengdu Zongheng Automation is one of the few companies in the country specializing in vertical takeoff and landing drones, with business segments including flight control and guidance products, UAV whole machines, and modules. The leadership was satisfied with my open-source flight control system design and maintenance capabilities, thus assigning me to control software and system development. During my time at the company, I participated in several research institutes and international projects (including simulation design at Mianyang Aerodynamics Research Institute, Harbin Institute of Technology's one-station-four-machine formation system, Oman national customer). Especially during the Oman national customer training, I excellently completed a month-long English training session (main scope: flight control product architecture, onboard flight software and flight SDK development platform, flight control simulator, etc.), receiving unanimous praise from the customer.