Implemented reinforcement learning (RL) techniques including Proximal Policy Optimization (PPO) and Adversarial Imitation Learning to train humanoid robots in tasks such as walking, jumping, and kicking, resulting in improved stability, accuracy, and performance.
Designed and developed a cutting-edge drone for nuclear safety inspections. Equipped with CV object detection (YoloV5), the drone can read critical data from radiated zones, while its visual SLAM enables accurate localization, RRT* allows for obstacle-free path planning and precise way-point motion planning for navigation through these zones.
The Unscented Kalman filter uses odometry and computer vision to calculate an optimal estimate of the robot’s location and navigate accurately to the destination, with tuning of the covariance parameters based on keeping the ground truth within the 1 sigma bound.
Developed a computer vision software using Python, yolov5, and ROS for object detection.
Developed a computer vision software using Python, OpenCV, and ROS for soccer field line detection. Utilizes a point cloud approach for improved accuracy, with 2D points transformed to 3D points in the robot’s frame of reference, allowing for definition of field lines with known thickness.
Developed Various Components necessary for a self driving car
Developed a convolutional neural network for classifying musical notes from spectrogram images with 90% accuracy.
Designed and created an innovative autonomous flower pot that uses computer vision and environmental sensors to track and move toward optimal sunlight and water locations.
Programmed an android app using IBM Watson API to deliver a satisfying user experience for connecting top applicants with top companies.
Programmed an android app using IBM Watson API to deliver a satisfying user experience for connecting top applicants with top companies.
FRC 2017 Robotics competition.