Allen Zhou

Founding Engineer @ Verne Robotics

About

I'm a robotics and control engineer with a strong focus on mechatronics, system dynamics, real-time actuation, and computer vision integration. I specialize in building intelligent robotic systems that combine dynamic modeling, control strategies, and sensor fusion to achieve precise and adaptive behaviors in real-world environments. In the field of ADAS, I’ve developed end-to-end vision pipelines using YOLOv5, YOLOv8, and YOLOX for real-time object detection, multi-object tracking, and motion prediction of road participants. These models were deployed on onboard computation units to enable low-latency inference and robust performance across varied lighting and weather conditions. In parallel, I’ve led the development of a tendon-driven soft robotic manipulator with tunable stiffness, where I formulated an optimization-based control strategy grounded in system energy models and geometric constraints. This included forward kinematics computation, multi-mode workspace analysis, and precise stiffness modulation using hierarchical PID control across embedded processors. Technical strengths: Mechatronics: Arduino, ESP32, CAN bus, ToF/VIVE sensors, actuator integration Control & Optimization: PID, LQR, MPC, passivity-based control, constrained optimization Perception Systems: YOLO, IMU fusion, embedded system integration Mechanical Design: SolidWorks, fast prototyping Software Tools: Python, MATLAB, ROS2, Simulink, Git, Arduino I'm actively pursuing roles in robotics, automation, and control engineering. I’m also excited to connect with engineers, researchers, and organizations advancing the future of autonomous machines.

Country

United States

City

San Francisco

Industry

Information Technology & Services

Skill

Computer-Aided Design (CAD), Product Development, Path Planning, Modeling and Simulation, Mechanical Product Design, Robot, Motion Control, Robotic Design, Kinematics, Automotive Engineering, Numerical Analysis, Analytical Skills, Mechanical Analysis, Computer-Aided Engineering (CAE), Robotics, Mechatronics, Python, SolidWorks, Arduino, Matlab

Experience

Founding Engineer

Verne Robotics

2026-1 - Present · 9 mos

San Francisco, California, United States

Robotics Automation Engineer

Adas Eco

2024-5 - 2025-9 · 1 yr 5 mos

Irvine, California, United States

Built and maintained a cloud-based training workflow on SSH-connected GPU servers, enabling stable, resumable model training with real-time metric tracking and automated logging. Designed and deployed YOLO-based perception pipelines (v5/v8/X) for autonomous-driving applications, reaching 98% detection accuracy across diverse weather and lighting conditions. Integrated multi-object tracking with IMU fusion to assign persistent IDs and predict trajectories of vehicles, pedestrians, and the ego-vehicle for improved situational awareness. Implemented collision-time estimation by combining object prediction and IMU data to issue early-warning pre-alerts for high-risk interactions. Deployed optimized YOLO models on TI J722SXH01 embedded hardware, achieving 25 FPS at 720p with ~40 ms latency via hardware-accelerated inference. Developed self-localization for autonomous forklifts using camera + LiDAR fusion, AprilTag-based positioning, A path-planning, and SLAM mapping* to enable collision-free navigation in warehouse environments.

Graduate Research Assistant

University of Pennsylvania

2024-5 - 2025-5 · 1 yr 1 mo

Philadelphia, Pennsylvania, United States

Engineered a tendon-actuated, tunable-stiffness soft robotic arm with both local (spring-based) and global (tendon-driven) actuation for adaptable compliance. Designed, prototyped, and integrated modular coiled-spring actuators with real-time stiffness variation up to 10×, deployed in a 3-DOF soft manipulator segment. Performed workspace modeling and geometric analysis to evaluate task feasibility across multi-tendon actuation modes. Developed a Lagrangian energy-based optimization framework for quasi-static stiffness control; validated results with OptiTrack motion-capture experiments. Built a hierarchical multi-segment control policy using CAN-bus communication and PID feedback, reducing end-effector error to < 2 % during testing.