Engineering Portfolio
Robotics & Control Systems — Carnegie Mellon University
I design and implement feedback control, estimation, and motion planning algorithms for autonomous vehicles and legged robots. My work focuses on turning mathematical models into robust, real-time systems that move safely in uncertain environments.
Ultimately, I am interested in using robotics to make transportation and logistics safer, more accessible, and less labor-intensive, especially in challenging or high-risk scenarios.
About Me
I am a graduate student in Mechanical Engineering at Carnegie Mellon University, focusing on robotics and control. I enjoy working at the intersection of modeling, estimation, and real-time control, where theory has to survive noisy sensors, hardware limits, and tight deadlines.
Recently, I have been working on autonomous vehicle tracking using EKF and LQR, remote tele-operation platforms for logistics during epidemics, gait transition control for quadruped robots, and pneumatic-based flexible suspensions. I am especially motivated by problems where better control and autonomy can directly improve safety or human workload.
- Technical interests: State estimation (EKF/SLAM), optimal control (LQR/MPC), reinforcement learning for locomotion, mechatronic system design
- Tools: ROS/ROS2, Gazebo, MATLAB/Simulink, Python, C++, Git, LaTeX
- Projects — four featured projects
- Resume — 1-page PDF
- Justification Memo — design rationale
Projects & Experiences
Each project below follows an IMRD / IDEC-style structure, highlights social value, and clarifies my specific role. Click a project to open a full-page description.
Vehicle Autonomous Tracking with EKF, LQR & SLAM
CMU 24-677 — Advanced Control Systems Integration
Introduction / Problem. The goal of this course project was to enable an autonomous ground vehicle to track a reference path in an urban-like environment using noisy sensor data. Reliable localization and robust lateral control are critical for reducing accidents and improving traffic efficiency as autonomous vehicles become more common.
Method / Design. I modeled the vehicle using a nonlinear bicycle model and implemented an Extended Kalman Filter that fused wheel odometry, IMU, and range measurements into a consistent state estimate. On top of the linearized dynamics, I designed an LQR controller for lateral and heading tracking and integrated a graph-based SLAM module so that the controller could operate in a partially unknown map.
Results / Evaluation. In simulation, the controller achieved stable tracking with a lateral error under 0.15 m RMS on curved roads and recovered gracefully from injected sensor noise and small disturbances. Compared to a baseline PID controller, the LQR design reduced overshoot and improved settling time, especially when the map was updated online via SLAM.
My Contribution / Role. I implemented the full estimation–control pipeline, including the EKF derivation, discrete-time LQR design, and ROS-based integration. I also generated the evaluation metrics and plots used to compare different controller tunings.
Remote Driving Contactless Logistics Platform for Epidemic Scenarios
Shanghai Jiao Tong University — Team Project
Introduction / Social Value. During severe epidemics, front-line logistics workers face high exposure risk. Our project proposed a contactless logistics platform that allows an operator to remotely drive a small electric vehicle to deliver goods in hospitals or quarantine zones, reducing unnecessary person-to-person contact.
Design / Method. We built a ROS-based teleoperation system that streamed multi-view camera feeds and LiDAR data to the operator, overlaid predicted trajectories, and provided a visual assistance HUD. A joystick interface translated the operator's commands into low-level motor and steering signals, while safety modules enforced speed limits and soft geofencing.
Results / Evaluation. In field tests on a mock hospital corridor, the platform completed delivery routes with centimeter-level tracking accuracy and maintained stable video and control latency within 150 ms. The system was recognized at a national-level innovation competition for its potential to support safe logistics in public health emergencies.
My Contribution / Role. As team lead, I coordinated system architecture, implemented the ROS nodes for trajectory prediction overlay, and handled integration between the vehicle's embedded controller and the remote user interface. I also led the testing protocol and documented the performance metrics.
Seamless Gait Transition via Optimal Control & Reinforcement Learning
Quadruped Locomotion Research
Problem / Motivation. Legged robots often need to switch between different gaits (for example, from a slow walk to a faster trot) when terrain or speed requirements change. Abrupt transitions can lead to loss of balance or large tracking errors. The goal of this project was to design a controller that produces smooth, dynamically consistent gait transitions.
Idea & Design. I combined model-based optimal control with reinforcement learning. First, I used trajectory optimization to generate reference motions that satisfied the robot's dynamic constraints. Then, I trained a policy that learned to track these references and interpolate between them, using a reward that balanced tracking accuracy, energy efficiency, and foot-slip penalties.
Evidence / Results. In simulation, the controller achieved smooth transitions between walking, trotting, and bounding at different speeds, with reduced body pitching and improved foot clearance compared to a purely heuristic state machine. The learned policy generalized to moderate modeling errors and small terrain irregularities without re-tuning.
Contribution / Role. I implemented the dynamics interface, designed the reward function, and evaluated performance using center-of-mass tracking error and contact consistency metrics. I also generated visualizations and videos to analyze how the learned policy blended between optimized gait primitives.
Pneumatic Artificial Muscle-based Flexible Suspension
Mechatronics & Vibration Control Design
Introduction / Objective. Traditional passive suspensions have limited ability to adapt to different road conditions or payloads. This project explored a flexible suspension concept using pneumatic artificial muscles (PAMs) as tunable, compliant actuators to improve ride comfort and vibration isolation.
Design / Method. I modeled the quarter-car dynamics with PAM elements and designed a feedback controller that adjusted chamber pressure based on measured acceleration and displacement. The prototype included pressure regulators, displacement sensors, and an embedded controller that executed the control law in real time.
Results / Evaluation. Bench-top experiments with sinusoidal and step-like road profiles showed a noticeable reduction in transmitted acceleration compared to a baseline passive spring-damper system. The system could trade off stiffness and comfort by changing the pressure set-points, demonstrating the potential of PAM-based suspensions for small vehicles or assistive devices.
My Contribution / Role. I led the modeling and control design, implemented the real-time controller, and performed system identification to obtain PAM stiffness curves. I also designed plots and frequency-response visualizations to communicate the performance improvements.
Resume
This portfolio is accompanied by a concise one-page resume tailored to robotics and control roles. The resume highlights technical skills, research experience, and relevant industry internships.
View 1-page Resume (PDF)Contact
- Email: jingboz2@andrew.cmu.edu
- LinkedIn: https://www.linkedin.com/in/jingbo23zhang/
- GitHub: https://github.com/jbzhang23
Justification Memo
For this course, I prepared a separate justification memo that explains how this portfolio is targeted to employers in robotics and control engineering. The memo discusses my audience analysis, project selection, and document design decisions (navigation, visuals, typography).
- Why these four projects best represent my skills in modeling, control, and real-time implementation
- How the project descriptions emphasize impact, social value, and my specific role
- How the layout, headings, and visuals support quick scanning by technical recruiters
(If required, upload memo.pdf to the repository so this link works.)