Role: Team Leader, Software & Electronics Engineer

This platform was developed across three competition seasons and evolved from a simple vision-guided line follower into a tightly integrated multi-sensor rescue robot. The architecture separates high-level perception on a Raspberry Pi from real-time motor control and sensor polling on a microcontroller, ensuring deterministic actuation while running computationally heavy vision pipelines. Please refer to the Technical Design Paper for the best possible information. The information on this website is meant to be a brief overview only.

Competition Results

The robot achieved:

• National Champion in RoboCup Rescue Singapore 2022
• Champion in RoboCup Asia Pacific 2022
• Best Educational Value in RoboCup Asia Pacific 2022
• Judges' Award in RoboCup Asia Pacific 2022
• National Champion again in RoboCup Rescle Singapore 2023

Perception and Vision System

Line tracking algorithm

The primary navigation modality is camera-based line tracking using OpenCV. The pipeline performs adaptive thresholding followed by morphological filtering to remove noise and contour extraction to isolate the line geometry. This allows the robot to maintain stable tracking through intersections, gaps, and partial occlusions. A secondary top-down camera was later introduced to improve victim detection reliability and provide additional spatial context when entering the rescue zone.

Lighting variation was a major source of failure in early iterations. To address this, a custom reflectance sensor array was mounted above the camera and fused with the vision estimate. This provided a lighting-invariant signal that stabilised the controller when the camera exposure was saturated or when the line contrast was reduced by shadows.

Control Architecture

Line following is implemented using a PID controller operating on a fused error signal derived from both vision and reflectance data. A finite state machine governs transitions between line tracking, obstacle avoidance, rescue zone behaviour, and return navigation. Running this logic on a Teensy 4.1 ensured consistent loop timing and eliminated latency spikes observed when control was executed on the SBC.

Electronics and PCB Evolution

Body Plate PCB Schematics

Body Plate PCB Layout

The electronics stack evolved from a through-hole prototyping board to a compact surface-mount PCB integrating power distribution, sensor breakout, and microcontroller support. Moving to SMT reduced footprint and improved electrical reliability while enabling a modular stacked architecture that allowed rapid iteration between competition seasons.

Bottom Plate PCB Layout

Mechanical Design

A custom gearbox was designed to drive four wheels using two motors, reducing overall width while maintaining torque for obstacle traversal. Omniwheels were later introduced to stabilise turning on rectangular speed bumps, reducing lateral slip that previously caused controller oscillations. The mechanical layout was optimised for narrow entrance navigation and repeatable wall tracking in the rescue zone.

Robot Evolution

1. 2022 National Champions

• Using OpenCV on Raspberry Pi 4B to detect and navigate a line and obstacles.
• PCB (only THT) custom ordered to act as a base.
• Arduino Mega for motor control and sensor readings.
• Mechanical design using Fusion 360 and 3D printing.

2. 2022 Regional Champions

• Upgraded to Teensy 4.1 as microcontroller
• Iterated on OpenCV method by exploring different functions.
• Created and added a light sensor array PCB on top of the camera to improve line detection in different lighting conditions.

3. 2023 National Champions

Model of the final robot

• PCB revised to include SMT components.
• Another camera (top down) was added to improve depth perception and victim detection.
• 3D Printed gearbox for a mechanical 2 motor but 4 wheel drive was created to reduce footprint and improve reliability in navigating through narrow spaces. (e.g. entrance).
• Created and 3D Printed Omniwheels to improve turning on the rectangular speed bumps.
• Added more TOF sensors for better victim, obstacle detection and wall tracking in the rescue area.