Phat C. Vo

These robots, equipped with steering-based locomotion mechanisms, have been deployed for security patrol tasks in public areas and construction sites, as well as for logistics operations in industrial environments. They autonomously navigate complex, shared internal roadways populated by humans and other vehicles such as trucks and forklifts. This project is also designed based on Conflict-Aware Dispatcher for deployment in large factory environments where AMRs operate on shared internal roads (indoor + outdoor). The system enables coordinated operation among robots and seamless interaction with web-based tools and human operators. It serves as the central control layer for scalable, safe, and intelligent fleet management—laying the foundation for smart factory automation.

I was responsible for several core features, including:

• Build and maintain a comprehensive simulation environment for testing and validating algorithms.
• Behavioral Planning for real-time decision-making, including obstacle avoidance and speed profile planning.
• Global Navigation, tailored to scenario requirements and factory map data.
• Local Trajectory Generation Algorithms to ensure smooth and feasible motion control.
• Collision Risk Assessment and Safety Validation to guarantee reliable and safe navigation.
• Task Scheduling and Dispatching: Dynamically assigns schedules and inspection tasks to each robot based on real-time conditions and system priorities.
• Multi-Robot Path Planning and Conflict Resolution: Manages priority-based rerouting and waiting strategies to prevent congestion and resolve conflicts in real time, especially in narrow roads and intersections shared by multiple robots.
• Monitoring and Visualization: Collaborated with the front-end team to integrate a web-based dashboard that supports real-time, bidirectional messaging and logging database. It provides live updates on robot statuses, and task progress, enabling operators to remotely supervise the entire fleet.

BEEBOT2 is an autonomous mobile robot taken from an empty workspace to a machine that drives, sees, maps, localises and navigates — built in public as a written build log alongside a companion video series.

The series is deliberately not organised as a tour of individual packages. It follows the robot: it learns to drive, to take commands, to feel its own motion, to see, to build a map, to know where it is, to plan a route, to protect itself, to charge, to do a job, and finally to work alongside other robots. Each article starts from a problem the robot cannot yet solve, builds the smallest thing that solves it, measures the result, and ends with the next problem.

The build covers:

• Drive system — two BLDC gear motors on an MDROBOT MD200T controller over a single RS485 bus, wrapped in a ros2_control hardware component.
• Digital robot — URDF/xacro description with geometry regression tests, and a Gazebo Harmonic warehouse that mirrors the real launch graph.
• Sensing — LiDAR, a MicroStrain 3DM-GX5 IMU, wheel encoders, RGBD camera and bumper, fused into one motion estimate with robot_localization.
• Power — a lithium pack with DALY BMS telemetry published as sensor_msgs/BatteryState, so the stack can reason about remaining runtime.
• Autonomy — slam_toolbox for mapping, map-quality scoring tools, AMCL for localisation, and Nav2 for planning and control, with twist_mux arbitrating command priority.
• Safety and service — protective and warning fields, E-stop integration, failure-mode analysis, docking and charging, payload HMI, and a robot API built for fleet attachment.

The most important architectural decision is that the simulation is not a side project: sim.launch.py and robot.launch.py are deliberate mirrors, sharing the same controller configuration, the same diff_drive_controller and the same EKF. Only the hardware component underneath swaps — Gazebo on one side, the RS485 driver on the other. The same code also runs a 300 kg warehouse platform, which is the cheapest possible proof that the stack is not secretly tied to one chassis.

I worked with the Educational Robot Platform (ERP) developed by WeGo Robotics, which was integrated into the MORAI simulation environment for autonomous driving research and development. Using this platform, I implemented and tested core motion planning components, specifically:

• Implemented local planning and motion tracking algorithms in the MORAI simulator.
• Integrated locally developed algorithms into the server-based simulation environment via the bridge node.
• Developed a hybrid control framework for concurrent simulation and real robot operation, using simulation outputs to drive the physical robot.

To support development and evaluation, I published key planning and tracking data to ROS topics, allowing for real-time visualization, logging, and further analysis. This setup provided a practical and extensible environment for experimenting with and validating autonomous driving algorithms on a compact, education-oriented platform.

A small Qt-based demo implementing RRT* with Dubins paths.The GUI supports drawing rectangular obstacles interactively. To draw an obstacle: press the left mouse button at one corner, drag to the opposite corner, then release. The rectangle will be added to the obstacle list and the scene will refresh.

The included launcher script simplifies running the app: chmod +x run_app.sh && ./run_app.sh

Human-Robot Teleoperation (HRT) is the focus of my Ph.D. thesis project, developed using MATLAB and supported by a series of published research works. It aims to establish a reliable and responsive teleoperation framework that balances both human control and robotic autonomy. The project addresses four key aspects:

• Robustness: Stable performance despite uncertainties from the operator, environment, or sensors.
• Performance: Effective task execution with speed and accuracy.
• Perception: Enhancing the operator’s awareness of the remote environment.
• Transparency: Accurately reflecting the environment and robot state to the operator.

To find further details in my thesis defense presentation available in the Education section.

This project designs and controls a Variable Stiffness Actuator (VSA) for quadruped robot legs to enhance adaptability, shock absorption, and energy efficiency. The VSA consists of two actuators: one controls joint position, the other adjusts stiffness by modulating an elastic element. This allows real-time tuning of joint compliance independently from position.

The control system includes:

• A high-level controller that plans motion and desired forces,
• A low-level controller that adjusts both position and stiffness using sensor feedback.

By integrating VSA into each leg, the robot can achieve softer landings, better stability on uneven terrain, and improved safety when interacting with humans. This approach mimics animal-like locomotion and is ideal for dynamic and unpredictable environments.