Projects
I come from an engineering background which has sort of molded a "product is king" kind of thinking but my professors at Columbia have helped me see why that does not work.
I have worked on multiple projects in the past around AI and robotics but if they are not documented well in a manner that can be shown in a succinct and simple manner in which everyone can understand than frankly my dear, no one gives a damn. Implementing an interesting research paper or working on an exciting project is great but I have also started to pair it with an equally good way to present it. This makes it complete, and if you are reading this, I wish you learn from my mistakes
Publications
Decision Making for Human-in-the-loop Robotic Agents via Uncertainty-Aware Reinforcement Learning
In the reinforcement learning paradigm, an agent is able to act mostly autonomously in solving a task, but can request help from an external expert when needed. However, knowing when to request such assistance is critical: In this paper, we present a Reinforcement Learning based approach to this problem, where a semi-autonomous agent asks for external assistance when it has low confidence in the eventual success of the task. The confidence level is computed by estimating the variance of the return from the current state.
​
Reinforcement Learning for Inventory Management
Warehouses with large stock quantities make most operational decisions manually. These procedures can be optimized and automated using AI and we present a simulated market environment that incorporates the behavioral patterns of a small business and a reinforcement learning method that can be used to identify optimal order quantities and optimal order times.
Robotics and autonomous systems
Take a Look
Model Predictive Control
After a path of waypoints has been selected for a vehicle to drive on, we need to convert it into input signals that a vehicle can accept. An MPC algorithm optimizes the steering angles and acceleration such that the driven path is as close to the desired path. Here I have used C++ to implement MPC on a driverless vehicle simulator.
CCTV controlled robot planning in warehouses
Built a wireless mobile robot as a prototype for warehouse robots that can be controlled by CCTV cameras. An overhead camera uses computer vision to analyze the robot's position and obstacles to plan the shortest path to an end goal. This system reduces the need for expensive onboard sensors and localization systems.
Skills used: Raspberry Pi, Mechanical Design, Computer Vision, A* path planner
Sensor Fusion
Autonomous vehicles use sensor fusion to predict the speed and position of vehicles around them. Kalman filtering is used to fuse data from multiple sensors. In this project I programmed an EKF and UKF using C++ that fuses measurements from radar and lidar sensors in a simulator to accurately predict a car's state.
RRT Planner
Implemented RRT Planner in Python for a 2 DOF robot arm
Machine Learning
Take a Look
Vehicle detection using SVMs
It will be stupid to detect vehicles using SVMs. Trust me, I've done it. I had to extract HOG features for every image in my vehicle dataset. Then during inference, I predicted the probability of a car on around 1000 possible bounding boxes. Add a few more layers of complication and you get an inefficient vehicle detector. I still managed to get it to work and if you'd like you can read the GitHub repo to know how I did it in Python.
Naive Bayes Classifier
Wrote a naive bayes classifier from scratch because sklearn just isn't as much fun. The classifier uses data of positions and velocities of cars and predicts whether a car will continue straight or take a left or right turn.
I have added some of my notes as well to understand it better
Computer Vision
Take a Look
Gradient Visualization
This is your Project description. Whether your work is based on text, images, videos or a different medium, providing a brief summary will help visitors understand the context and background. Then use the media section to showcase your project!
Deep Learning
Take a Look
GANs
This is your Project description. Whether your work is based on text, images, videos or a different medium, providing a brief summary will help visitors understand the context and background. Then use the media section to showcase your project!
CycleGAN
Converting images from one domain to another. In this case from horses to zebras and vice versa
