Highway Driving is a task that requires long-term and short-term planning. We developed an End-to-End Reinforcement Learning (RL) approach to achieve this goal. My contributions are mostly to reward function design and multi-simulation training archituecture.

If you want to see more. Github - Paper

• Imlemented an OpenAI Gym Interface for CARLA
• Designed a multi-simulation training framework with Docker
• Trained and Tuned TRPO, PPO2, DDPG and A2C Agents on Kubernetes
• Experienced with Frenet Trajectory Planning and Low Level Control (PID, Cross-Track Error)
• Increased Speed, Comfort and Safety Metrics and Published Contributions in IEEE CASE 2021

Well, This project started with me wanting to know who came to my wedding. I implemented a framework which you can extract unique IDs from a video using OpenCV and Dlib Libraries. I spent quite amount of time for debugging and making sure that I catch everyone showed up :D.

If you want to see more. Github

• Completed a demo that can identify Unique IDs from the Video
• Got a further understanding of Cascading Classifiers, HOG and implications of CNN Layers
• Implemented Affine Face Alignment (Centering, Rotationg, Scaling) from Face Landmarks
• Experimented with Clustering Algorithms (DBSCAN)

Yeap, This is how i gained interest in AI. I gathered a team to particiate in RoboSub2015. Goal was to navigate underwater and complete pre-given tasks. This was a full-stack hardware and software system design and implemetation.

If you want to see more. Design Specifications

• Lead a Team that Designed and Built 6-thruster Underwater Vehicle
• Implemented Perception Algorithms to Detect Shapes and Digits to Complete Tasks
• Designed Mission Planner and Trajectory Planning Algorithm
• Experienced with Kalman Filter and Sensor Fusion for state estimation
• Designed a GUI and server-client application (C++) over TCP/IP for monitoring, testing and tuning purposes
• Successfully Completed Qualifying Rounds and Competed in Finals