2017-2019

  1. Self-Driving Car: Mahindra Rise Challenge. This competition involves developing a Driverless car for Indian traffic scenarios. My work in this project involved the development of both localization and planning pipeline for the autonomous car. For the motion planning pipeline, I extended the time-scaling concept to derive a closed-form analytical solution for performing complex maneuvers like merging onto the ongoing traffic by considering the collision risk. The successful implementation of the same on the autonomous vehicle and formed the crux of my thesis topic and led to publications in IV (Intelligent Vehicle Symposium-18) and AIR (Advances in Robotics-19). For the localization pipeline, I also worked on the sensor calibration, sensorfusion, and state estimation modules for developing the SLAM framework of the autonomous car.

2017-2019

  1. Multi Agent Systems This work mainly involved development of both deterministic and stochastic variants of egocentric version of the famous velocity obstacle(VO). The idea is an egocentric velocity obstacle helps in improving real time implementations of collision avoidance in dynamic environments as there is no dependency on state estimation techniques to infer the robot pose and velocity. My contribution to this work include reformulating the the velocity obstacle to adapt to an egocentric framework, conducting the real time experiments on Bebop drone and writing the papers which got published in AIR (Advances in Robotics-19) and ROMAN(International Conference on Robot and Human Interactive Communication 2019).

2019-2020

  1. Investigation of non-parametric uncertainty in Motion planning In this work, we developed an efficient algorithm for solving a class of chance constrained optimization by representing the non-parametric uncertainty as functions in Reproducing Kernel Hilbert Space(RKHS). My contributions in this include both developing an computationally efficient implementation of the proposed idea in C++ and its deployment on Bebop drome to conduct real time experiments. The efforts for this work got published in RA-L with ICRA 2020

2017-2018

  1. Risk Aware Merging. The main objective was to develop a risk aware merging behavior, for a traffic like scenario. I have developed a framework which has a two layer structure which ensures generating a collision-free merge maneuvers even in dense traffic scenarios. This lead to publications in IV(Intelligent Vehicle Symposium-18,19)

2016-2017

  1. Localisation and Navigation in GPS Denied Environment In this project I have developed an algorithm that fuses the sensor data from a visual sensor and an IMU to estimate the robot’s current location and navigate the robot to its destination with obstacle avoidance in GPS denied environment. Not having the GPS data makes the problems more challenging due to absence of fixed global reference frame. The planning stack was implemented using the RRT planner from the MRPT toolkit in Tory Parameter (TP) space, was deployed on Clearpath A200 mobile robot, and tested for its efficacy.

2017

  1. Road Segmentation with different Classifiers In this work we developed different classification models for road segmentation. The classification models we used are Superpixel based Classification(SVM, Naive Bayes, KNN, Random Forests,), Neural Network Based Classification (CNN) . This is implemented KITTI Road Datasets.