Selected Projects
Control Systems and Theory:
Fuzzy Logic based PID Controller for BLDC Motor speed Regulator:
Designed a Fuzzy Logic PID Controller for BLDC motor speed regulation from scratch, simulated the model using Simspace and Matlab.
Controller for a Crane System
To design controller and observer for the system which can control the loads suspended from the cables of the crane. The project flows in steps, starting from deriving the equation of motion of the system. Then, linearizing the system at equilibrium points, designing an LQR controller in SIMULINK, designing a Luenberger Observer for both linear an well as non-linear system. The calculations on non-linear system is computed using ode solver Ode45. Finally, designing a LQG controller using Kalman Filter and LQR Controller.
ROS 1 and ROS 2:
3D Mapping through an Autonomous Drone
Independently created a 3D mapping system for a quadcopter by integrating Navigation (NAV) and Simultaneous Localization and Mapping (SLAM). Authored the complete Drone Plugin for ROS2
Autonomous Navigation of Turtlebot3 using ROS and Gazebo.
Developed an Open Loop controller system designed to project the 2D pose of the robot over time. This system enables autonomous movement of the robot in the X-direction for a predefined duration of 10 seconds. This innovative controller enhances the robot's ability to navigate and execute tasks with precision.
Autonomous Bot
Designed and simulated an autonomous robot capable of mapping and self-localization in uncharted environments through Simultaneous Localization and Mapping (SLAM) technology. This versatile robot holds significant potential for applications in the military, logistics, mapping, and personal sectors."
Adaptive Monte Carlo Localization
probabilistic localization system for a robot moving in 2D. It implements the adaptive (or KLD-sampling) Monte Carlo localization approach (as described by Dieter Fox), which uses a particle filter to track the pose of a robot against a known map.
C++ :
Route Planning using A* on IO2D map
implemented the A* path planning algorithm on an OpenStreetMap Io2D map. This algorithm efficiently computes the fastest path between two points specified by their X and Y coordinates while also providing the distance of the path.
Deep Learning :
Detection of Lower Grade Glioblastoma using UNET and densenet121-UNET
In my final year undergrad dissertation, I undertook a collaborative effort with fellow undergraduate researchers. Our focus was on the detection of Low-Grade Glioblastomas (LGG) utilizing the TCIA dataset. Employing state-of-the-art UNET and Densenet121-UNET models, we achieved remarkable accuracy rates of 98% and 97.5%, respectively.
2 D Pneumonia detecion on X-Ray dataset
crafted a machine learning model for pneumonia detection in the NIH X-Ray dataset. Leveraging the robust VGG16 and ResNet50 architectures, achieved impressive accuracy rates of 97.5% and 98%, respectively.
Depth Estimation for Stereo Images
Implemented to compute depth from Stereo Images. Tested on three different datasets, each containing images of the same scenario but from different camera angles. Followed process of Feature Matching using SIFT, RANSAC, Essential Matrix, Camera Pose, and Stereo Rectification
Autonomous Lane Detection for Cars
Utilized OpenCV, implementing Hough Transform to identify and map out the distinct lines of lanes on a freeway, enhancing lane detection accuracy. Applied Canny edge detection in OpenCV to precisely locate lane boundaries on a freeway, a crucial step for reliable lane tracking and navigation.