Yasantha Niroshan

Currently co-mentoring 2 ongoing project teams.

• Sim-to-Real Deployment Framework for RL-based Robots — Reusable framework that automatically converts robot policies trained in simulation (MuJoCo + PPO/Stable-Baselines3) into deployable code for ESP32-class microcontrollers, replacing today’s manual sim-to-real conversion process.
• Proactive Social Navigation for 2D LiDAR Mobile Robots — Navigation approach for 2D-LiDAR mobile robots that anticipates the motion of people nearby to plan socially aware, non-disruptive paths.

Mentored ~16 student teams through the full embedded product cycle — 5 of the teams below published at the ERU Symposium 2025.

• ASCILAM — Collaborative terrain-mapping system in which LiDAR-equipped M-Bot scouts stream scan and odometry data over Wi-Fi (UDP) to a Raspberry Pi coordinator that fuses everything into a global map in real time.
• DropToPrint — ESP32 module that retrofits stock 3D printers: G-code uploads over Wi-Fi straight to the printer’s microSD card plus remote serial control from a web UI — no firmware modifications or OctoPrint required.
• Respiration Monitor — In-mask embedded tag that measures CO₂ concentration, temperature, and humidity in real time, monitoring respiration patterns with smartphone connectivity.

  Paper · co-authored A smart mask for wireless, real-time monitoring of CO₂ and humidity — Introduces the embedded sensing device and alerting pipeline for detecting unsafe CO₂ and humidity build-up inside face masks.

• Shadow — Privacy-conscious wellness platform that fuses wrist-wearable, phone, and laptop signals through a hybrid CNN trained on WESAD for real-time TinyML stress detection.
• Smart River Water-Level Monitoring — Low-cost, energy-efficient water-level gauge with threshold alerts and cloud logging, built for remote riverbank communities where GSM-based flood warning is impractical.

  Paper IoT-based river water level monitoring system using LoRa technology — Presents the LoRa sensing-node and network design that delivers reliable early flood warning at a fraction of the cost and power of GSM alternatives.

• GuitarPal — Guitar-learning assistant that lights addressable LEDs under every string and fret and senses finger placement, giving beginners real-time visual guidance and feedback via a Flutter app.

  Paper Guitar-Pal: smart guitar learning assistant — Describes the interactive LED-fret guidance hardware aimed at the steep early learning curve behind first-year dropout among novice guitarists.

• CCTV IntelliGuard — Edge module that upgrades existing CCTV cameras: OpenCV motion pre-detection gates a quantized, pruned YOLOv8 on a Raspberry Pi, storing timestamped clips and pushing mobile alerts via FCM.

  Paper CCTVIntelliGuard: an intelligent edge-based human motion detection and alerting system for CCTV surveillance — Details the hybrid OpenCV + YOLOv8 pipeline that adds real-time human detection and alerting to conventional surveillance systems.

• CTS-Guard — Wristband-and-ring wearable that uses IMU sensors to track wrist posture and repetitive finger movement for early detection of carpal tunnel syndrome risk.

  Paper CTS-Guard: a wearable smart device for prevention of carpal tunnel syndrome — Presents the posture-monitoring wearable that flags risky wrist positions before median-nerve damage develops.

Mentored ~8 student project teams building autonomous robots on the Kobuki platform with ROS 2.

• Kobuki Robot Control System — Complete ROS 2 control stack for a Kobuki base with LiDAR and Kinect — EKF and SLAM Toolbox for localization and mapping, plus a React web UI for teleoperation and autonomous navigation over rosbridge WebSockets.
• Kobuki LiDAR + Kinect Semantic Robot — Semantic mapping robot that fuses YOLO detections with Kinect depth and LiDAR range to pin recognized objects on a SLAM map, navigating with custom A* and taking voice commands over ROS 2 DDS.
• Vision-Based Leader–Follower Platooning — Autonomous platooning system in which a Kobuki QBot follows a remote-controlled leader car using AprilTag detection, Kinect depth sensing, and PID control, with LiDAR-based localization and a safety monitor.
• BotZilla — Object Detection, Collection & Placement — Autonomous robot that detects, collects, and places cubes in an indoor arena using YOLOv8-nano vision and RGB-D sensing, running on a Raspberry Pi 4 on the Kobuki QBot platform.