My Projects
Collaborated in a team to design and develop an advanced assistive robotic arm for individuals with upper limb impairments, concentrating on precise mouth tracking and adaptive utensil handling. Led the development of essential modules in ROS, incorporating Google’s MediaPipe for real-time facial landmark detection and utilizing depth cameras for accurate 3D spatial mapping.
In this project , I developed a navigation stack for a car by creating Python drivers for sensor calibration using GNSS puck and VN-100 IMU , utilizing ROS . To enhance localization accuracy, I formulated a sensor fusion algorithm that combined GPS data with IMU dead-reckoning estimates, resulting in improved error estimation for yaw, acceleration, and velocity. Leveraging MATLAB for this process, I achieved over 90% accuracy in the final implementation.​
Worked as a part of the team to implement a personalized ETA tool that utilized sensor fusion from GPS and IMU sensors, resulting in a 10% improvement in travel time accuracy compared to Google Maps estimates, with potential applications in urban planning. During the project, I conducted data collection by attaching an IMU to a bicycle, capturing speeds ranging from 0 to 3 m/s, and reduced noise in velocity data by 15% using median filtering on 1-second samples. I also proposed improvements by incorporating GPS for velocity calculation and accounting for pedestrian traffic to further enhance the model’s accuracy.
SELF RAISING PILLOW FOR CONGESTIVE HEART FAILURE PATIENTS
I spearheaded the design of an affordable, self-raising portable pillow for Congestive Heart Failure patients, utilizing an Arduino Uno, MAX 30100 sensor, linear actuator, DC motor, and L298 motor driver. This innovative pillow reduced the overall cost by more than 50% compared to existing market solutions, making it a more accessible option for patients.
In the fast-paced world of autonomous driving, understanding motion is crucial. Our project explored Detection and Tracking of Moving Objects (DATMO) using Optical Flow in CARLA, a high-fidelity simulation environment. We developed a pipeline that transformed raw LiDAR data into motion vectors, allowing autonomous vehicles to track dynamic objects with improved precision. By integrating Kalman filters and clustering techniques, we reduced velocity estimation errors, achieving 94.3% recall. This study highlights the strengths of Optical Flow and challenges traditional approaches, paving the way for more reliable perception systems in self-driving cars.