Academic work & credentials

To transition the project from reactive troubleshooting to structured systems engineering, I first reverse-engineered a digital twin in SolidWorks. Utilizing a Decision Analysis and Resolution (DAR) matrix, I replaced the friction drive with a reverse-tooth belt engagement strategy (using a T5 timing belt) to establish a positive mechanical interlock.

For the control architecture, I integrated an Omron NX1P2 PLC with a Weintek HMI, establishing a sequential state machine for deterministic motion control. I synchronized dual Cool Muscle servo motors and integrated Festo pneumatic cylinders to provide a rigid, physical "hard stop" at the target locations.

During physical integration, I successfully navigated complex hardware challenges, including resolving PNP/NPN sensor polarity mismatches, bridging communication protocols, and managing facility space and limited pneumatic air supply constraints.

The re-engineered system successfully eliminated all belt slippage and tooth jumping. By combining the PLC's velocity control with the pneumatic mechanical stoppers, the conveyor achieved a strict, repeatable positional accuracy margin of ±0.2mm across all target locations.

The prototype passed all Verification & Validation (V&V) criteria, successfully completing a 20-cycle automated repeatability test. Operators were provided with an intuitive HMI dashboard for seamless forward, reverse, and stop controls.

By effectively managing resources and repurposing internal company hardware (such as the Cool Muscle motors and Festo cylinders), the fully functional industrial testbed was delivered for approximately $130 — achieving all objectives well under the $500 budget.

To ensure the Omron robot could accurately fold the boxes, I designed and fabricated a custom prototype jig. The jig simulated the exact folding process, allowing me to perform physical adjustments and motion optimizations before full implementation.

The software architecture was built on ROS 2, where the box folding procedure was meticulously programmed to synchronize the robotic arm's movements with the custom jig's geometry.

Multiple iterations and tests were conducted to refine the procedural code and physical tolerances. The final implementation resulted in a highly reliable, automated workcell capable of executing the complete box folding procedure continuously without manual intervention.

I served as the Research Lead and Prototype Designer. I generated schematics using Fritzing and reverse-engineered the body frame, gearbox, and extruder digitally in SolidWorks.

The physical build featured a movement system utilizing a 12V DC gear-box motor to drive the platform along the X-axis, while the extruder navigated the Y-axis via a linear displacement mechanism. The extruder assembly used TT motors to open/close the syringes and step-down motors as heating elements to bring the edible wax to 400°F (liquefied state).

The prototype underwent strict verification against the initial requirements matrix. The system successfully executed user input, navigated the extruder over the product, and heated the wax into a liquid state.

While the extrusion of the edible wax was only partially successful (the final output was not fully coated evenly), the core mechanics proved viable. The project successfully served as a proof of concept for direct-to-food automated labelling.

I set up the LIMO robot with a LiDAR sensor to collect environmental data and launched the gmapping ROS package to process the laser range finder and odometry data into a 2D occupancy grid map.

After manually driving the robot to map the maze, I used RViz to identify key coordinates and place spatial waypoints. Finally, I configured the move_base node in ROS to handle the path planning algorithms required to reach those targets.

By launching the ROS navigation stack and continuously monitoring the robot's progress in RViz, I was able to fine-tune the parameters for physical noise. The LIMO successfully navigated the physical maze autonomously, avoiding obstacles and reaching all plotted waypoints.

I engineered the navigation logic using the "wall follower" algorithm. Rather than relying on a pre-computed global map, the Turtlebot was programmed to make dynamic decisions at each waypoint based on immediate environmental feedback.

The algorithm instructs the robot to move forward until it detects a dead end or obstacle via its onboard sensors. It then performs a calculated rotation and attempts a different path, repeating this state-machine loop continuously.

The Turtlebot successfully traversed the self-designed maze. The script proved robust against unknown geometries, effectively utilizing the wall follower logic to consistently find the exit path without manual intervention.

I fabricated and assembled the raw mechanical and electrical components into a complete robotic chassis. To perform machine-level actions, I programmed the microprocessor using Spin—a specialized, low-level hardware programming language.

A key mechanical addition was designing and creating a custom gear mechanism and holder using workshop resources to mount, rotate, and toggle an onboard torchlight.

The robot successfully passed all functional tests. Directional movement scripts validated the robot's ability to navigate accurately in specified vectors.

I further expanded the project's scope by experimenting with and successfully showcasing rudimentary color tracking capabilities, allowing the robot to detect and follow specific color signatures in its environment.