Projects

Avoid the wall

Autonomous wall detection and avoidance using optical flow
Abhinav Modi, Prateek Arora

Given a wall of unknown position and orientation, we developed a feedback system based on dense optical flow estimation using Gunnar Farneback approach to align with the wall and fly above or below it.
 Code   Video 

MountainCar-v0

Deep Q-learning for the Mountain-Car-v0 in OpenAI gym
Abhinav Modi

Trained a neural network using pytorch framework to perform deep Q-learning for the continuous state, discrete action-space mountain-car problem in the OpenAI gym environment.
 Code 

HODM

Human Obstacle Detection Module
Abhinav Modi, Kartik Madhira

Using software development principles - Agile Iterative Process we developed a package using a pretrained YOLOv3 network to detect humans in a sequence of input images.
 Code 

MyAutoPano

Panorama stitching using Traditional Supervised and Unsupervised Approaches
Abhinav Modi, Kartik Madhira and Prateek Arora

In this project we have used Homography to warp multiple images and stitch a panorama using three different techniques- Traditional approach using feature matching and RANSAC, Supervised approach to predict a 4 point parametrization of Homography between two images and an Unsupervised approach to predict Homography from the 4pt parametrization without the presence of a ground truth.
 Code 

Augmented Reality Tags

AR Tag detection and tracking, to project a 3D Cube in a video sequence
Abhinav Modi, Kamakshi Jain and Rohan Singh

Blob detection is used to detect three encoded(ID and orientation) tags which is followed by the use of homography and then perspective warping to project a cube on each tag based on its ID and orientation.
 Code   Video 

PB-Lite

Edge Detection using a modified Probability of Boundary Method
Abhinav Modi

The algorithm uses DoG, Leung-malik and Gabor filter Banks to generate texture, brightness and color maps to improve the boundary detection results of Canny and Sobel baselines.
 Code 

WALLE 2.0

Forward and Inverse Kinematic Modelling of a Mobile Manipulator Robot: WALLE
Abhinav Modi

The main idea of this project was to model the Kinematics of a reduced DOF(Degree of Freedom) model of the robot: WALL-E from the original Disney PIXAR movie: WALL-E. The objective is to perform a pick and lift operation using both the arms together on a mobile base. This will enable the robot to pick a box full of supplies, kept at fixed distance which is more than one arms reach.
 Video 

Finding Shortest Path

A-star and Dijkstra algorithms for a point robot and a rigid robot.
Abhinav Modi

Implemented A-star and Dijkstra algorithms for a point robot and a circular rigid robot of arbitrary radius on a given map.
 Code 

BIOBOT

PLSS Rover capable of carrying one crew only
Abhinav Modi, Hanish Mehta and Rohan Singh

The BIOBOT is a conceptual model of a one-crew carrying, semi-autonomous Lunar-Rover. The rover follows the astronaut with an on-board Life Support System such that astronaut can carry out extra-vehicular activities like sample collection with ease.
 Video 

Cable Suspended Load from a Quadrotor

Trajectory Generation and Control of a Quadrotor with a Cable-Suspended Load: A Differentially-Flat Hybrid System
Abhinav Modi

*Equal contribution
 Code