Zidong Z.

•Owned field deployment of a robotics automation integration system, connecting robots, RMS, and PLC controls through API + Modbus TCP interfaces; designed core coordination algorithms and sensor integration to enable stable real-world commissioning. •Built AutoCAD-based 2D map assets and mapped them into RMS-compatible navigation primitives; delivered traffic-control and task-scheduling logic for safer, higher-efficiency robot flow in constrained aisles. •Delivered BEVFusion-based perception improvements for mapping: trained and fine-tuned a camera+LiDAR fusion model to generate BEV feature tensors, integrated semantic segmentation into the SLAM/map-update pipeline, and improved map update accuracy by ~10%.

• Assembled, commissioned, and fine-tuned multi-brand industrial robots — Fanuc, KUKA, and Universal Robot for automated automotive production line (Nissan, Volkswagen, and Toyota). Partnered with simulation teams to take robotic cells from scratch to full-rate production, covering TCP and mastering calibration, safety zoning, DCS, cycle-time tuning, and on-site programming. • Programmed Allen-Bradley ControlLogix 5000 systems from I/O list to validated operation: verified field wiring and sensor integration, wrote ladder logic for signal interaction and output, verified communication via Ethernet/IP, and modified FactoryTalk View HMI screens, cutting combined robot and PLC programming time by 20 %. • Delivered on-site and remote support for glass-sealing applications; taught and optimized adhesive-glue trajectories, performed line testing, used multimeters and oscilloscopes to verify end-effector signals, proximity sensors, and lights, and used Fanuc RoboGuide to retrieve legacy parameters and accelerate robot commissioning with 15% fewer error events.

Engineered the frame and structural components for blood pressure medical devices, ensuring robust and ergonomic designs tailored for medical environments. • Designed and optimized PCB layouts using Altium Designer for integration with pressure sensors, ensuring seamless functionality and reliability in blood pressure monitoring systems. • Assembled and wired pressure sensors to Arduino microcontrollers for precise testing and validation of blood pressure measurements, focusing on signal accuracy and system stability.

I delve into the realm of Computational Fluid Dynamics (CFD) using Ansys Fluent, uncovering inefficiencies in structural design. Harnessing the power of SolidWorks and 3D printers, I craft new components to optimize airflow, revolutionizing the efficiency of Ionizer's corona discharge from a mere 61% to an impressive 70%. Similarly, the carbon brush discharge soared from 56% to a remarkable 62%, heralding a new era of enhanced performance. Undeterred by convention, I embark on a journey of innovation, delivering a groundbreaking PID control system. Departing from traditional methods of controlling HVAC solely through fan speed, I introduce a novel approach. Now, users can seamlessly regulate room particulate matter (pm) value levels via an intuitive interface, offering unprecedented control and comfort.

I spearheaded the implementation of the A* algorithm and orchestrated seamless communication between scripts and Gazebo plugin nodes. Crafting a function, I empowered the system to pinpoint the mouse's position as the A* goal, charting a trajectory to navigate the robot precisely within the dynamic Gazebo world. Innovatively simplifying obstacle detection, I harnessed the power of a single-camera model. By extracting optical flow from video recordings before the robot, I decoded valuable insights, enabling precise collision timing analysis of obstacles. This intelligence fueled control signals to hardware components, optimizing our robot's response. In the thrilling realm of the Gazebo world, I meticulously tested our system's mettle. Through adept programming of Gazebo plugins, publishers, and subscriber nodes, I orchestrated flawless communication between our model and software components. This rigorous evaluation included a captivating comparison with the performance of a two-camera model, illuminating the superiority of our streamlined approach.

I developed a streamlined iteration of the Doodle Jump game tailored specifically to enhance agent training. By optimizing gameplay mechanics, I created a more conducive environment for training agents, resulting in significant performance improvements and increased training efficiency. In addition, I successfully implemented a deep Q-learning model employing two distinct policies. Leveraging K-cluster analysis, I identified potential disparities in the distribution of training data. Through strategic navigation, I guided the agent to explore seldom-encountered data points, thereby bolstering the model's stability by an impressive 20%. This proactive approach ensured a more robust and reliable training process, ultimately enhancing overall performance.