# Andrei Tchouprakov > Software Engineer at Waymo Location: Mountain View, California, United States Profile: https://flows.cv/andreitchouprakov ## Work Experience ### Staff Software Engineer @ Waymo Jan 2020 – Present | Mountain View, California ### Sr Perception Engineer, Level 5, Lyft's self-driving division @ Lyft Jan 2018 – Jan 2020 | Palo Alto, ca ### LIDAR perception engineer @ Aptiv Jan 2018 – Jan 2018 | Agoura Hills, California LiDAR extrinsic calibration, visual odometry, SLAM, pose estimation, LiDAR tracking. ### Chief Architect @ E4D Technologies Jan 2014 – Jan 2017 | Richardson, TX Conducted research and feasibility studies of new techniques and hardware in the field of 3d computer vision, such as digital fringe projection, plenoptics ( lightfiled cameras), coded apertures, OCT, high speed laser drivers (1 ns pulse), MEMS mirrors, TI DMD, SoC (Intel Joule, NVIDA Jetson, Xilinx Zynq), Sony Pregius CMOS image sensors, ON Semiconductor KAI CCD and Python CMOS image sensors. Proposed and implemented the prove of concept allowing color imaging to be added to current scanner by integrating an array of RGB LED into the removable tip. Worked with electrical team on final version of the tip and developed the final software. As a result company was able to release color scanning capabilities to existing user without requiring hardware modification. Patent application: US 15/065,542 Continued improving and optimizing current software, achieving 40% faster 3D scan processing and less nosier data, utilizing techniques such as bilateral range image filter, median range image filter, weighted hashed voxel grid, reimplemented kd-tree data structure using SMID (AVX2 intrinsics). Dramaticly reduce file sizes, by saving only coordinates differences where apropriate, turncating redundant precision in floating points and compressing. Updated code to be C++11 conformat and updated compiler to VS C++ 2015. Initiated CAPA after observing that some intraoral 3D scanners didn't hold calibration over time. Led root cause analysis which determined inadequate mounting methods for some optical components were used (not a kinematic mount, too much and too hard epoxy). Successfully closed CAPA in additions to stability of calibration better overall accuracy was achieved. As result QA accuracy acceptance criteria was tightened by 30%. ### Director of system architecture @ E4D Technologies Jan 2009 – Jan 2014 | Richardson, TX Led development of the company's 2nd generation intraoral 3D scanner E4D NEVO (later rebranded as Planmeca PlanScan) US patent 09364300. Architected and developed cross platform (Windows and OS X) scanning software: hardware interface (CCD driver control, LCoS control, EEPROM, temperature sensors, accelerometer, using I2C and SPI drivers); pipelined (Intel TBB) scanning engine 3D point cloud generator (OpenCL) and 3D point cloud registration and 3D scanning visualization (point based surface render(splats) using OpenGL 4.0 GLSL multi-pass deferred shading). Toolset used: CMake, C++ (Visual C++ 2010, XCode clang), Boost, Ogre3D, TBB, Eigen, OpenCL, SFML, MKL. Perforce, git Successfully led the engineering effort to obtain regulatory certification(electrical safety IEC-60601-1, EMC/EMI IEC-60601-2, laser safety IEC-60825-1 class 2 laser product). Overseen transfer to manufacturing and developed manufacturing/QA tools: Camera and projection calibration (bundle adjustment), imaging and projection focusing estimation, pass/fail criteria New scanner was a big success on the market, in the first few months company sold more systems than in all previous years combined. Proposed and conducted research then successfully led implementation of a hardware upgrade and software for the 1st generation scanner. Upgrade allowed for the double frame rate of system by changing the drive waveform of the MEMS mirror from open loop sinusoidal to the closed-loop saw-tooth. The main challenge was dealing with resonance frequency and its harmonics of the mirror (hitting resonance cause breakage of the mirror). Over 80% of the existing users decided to update theirs systems. Toolset: research: NI FPGA PCI card + LabVIEW FPGA module, MATLAB Simulink; implementation: TI C200 Delfino MSU. ### Principal Research Engineer @ E4D Technologies Jan 2007 – Jan 2009 Worked on various user friendly CAD tools for designing a dental restoration. Implemented half-edge data structure for storing triangle mesh, using that structure allowed to simplify and speed up multiple mesh processing algorithms. Led R&D efforts and transfer to production of hardware, firmware and software for detection of brakeage/ wear of grinding tool bit in dental CNC machine by monitoring harmonics of the spindle in real-time during using MEMS accelerometer, TI DSP and FFT. Initiated and led modification to PID parameters, current limits on motors and trajectory computation for CNC machine to achieve maximum performance and test machine limits, as result milling time has been shortened by 20%. Replaced all trapezoidal motion profiles to S-curve profiles in CNC machine achieving faster and smoother motion during homing/loading/initialization and etc. Worked on toolpath generator minimizing time/tool life, as result by controlling feedrate according to how much material is being removed time has been shortened and tool life increased, created customizable setting file to allow add new types of the ceramic materials without much effort. ### Software Engineer @ E4D Technologies Jan 2004 – Jan 2007 Evaluated multiple structured light 3D scanning methods for 1st generation intraoral scanner (CMOS,dual scanning mirror and red laser diode). Existing methods using stripes (Grey code, fringes, etc.), failed to perform well intraorally due to high translucency of enamel and high penetration of the red laser. As solution I created new pattern and algorithm using lines instead of stripes and novel line encoding scheme. Method has been proven robust and modification of it are still in use in all scanner's generations produced by company. Developed motion calibration, forward and invese kinematics for 5 axis (3 translation, tilt, rotation) stage for bechtop 3D scanner. Conducted paper search for method for surface generation from unorganized 3D point cloud. As result implemented method based on running inverse FFT on 3D voxel grid to obtain iso-surface, to reduce memory requirement made modification to algorithm allowing rectangular voxel grid instead of square (possible to have <100um voxel even for full dental arch data) and created special weighting scheme to achieve better hole filling for dental scans. ### Software Engineer @ Spatial Integrated Systems Inc Jan 2001 – Jan 2004 Worked on a design and implementation of a 3D point generation algorithm from the series of 2D images form a digital video camera.  Proposed and developed the sub-pixel edge detection algorithm, achieving overall accuracy improvement of system by order of 5 comparing to regular Canny edge detection from Intel IPP.  Designed and implemented a texture mapping algorithm from multiple camera images. ## Education ### Master's degree in Mathematics Irkutsk State University ### Master of Science - MS in Mathematics Irkutsk State University ## Contact & Social - LinkedIn: https://linkedin.com/in/atchouprakov --- Source: https://flows.cv/andreitchouprakov JSON Resume: https://flows.cv/andreitchouprakov/resume.json Last updated: 2026-04-12