Completed Projects (funded)

Crashing to learn, Learning to Survive: Planning in Dynamic environments via Domain Randomization

• Trained End-to-end model for short term collsion free navigaiton in dynamic environemnts using deep reinforcement learning

• Used domain randomization to learn robust control strategies

• Open sourced gazebo based training environment with complete codebase and pretrained models.

Conference PaperOpen AccessVideo Colab Notebook

Ongoing projects (funded)

PSG Autonomous Vehicle Team

• Implemented robust perception system by combining classical computer vision techniques with deep learning using kalman filtering

• Developed end-to-end motion planning models using pure control systems as well as learning based techniques

• Currently experimenting Reinforcement Learning techniques in the Carla simulator  

Project Page 

Agile indoor quadcopter control using learning based methods 

• Bachelor thesis project (Inspired by work done at the ETH Flying machine arena by Professor Raffaelo D'Andrea's group)

• Training policies using deep reinforcement learning for agile quadcopter maneuvers

•Trained policies will be evaluated on the Carazyflie 2.1 platform 

Selected Independent Projects

Deep Learning Library

• Developed a deep learning library from Scratch in Python. 

• Intuitive API similar to pytorch.

CodeDocumentation 

ROS Integrated 4 DOF Robot Manipulator

• Designed and fabricated 4 dof Aluminium Robotics arm. Interfaced to ROS Moveit to implement Inverse Kinematics                                                                            

• Interfaced to ROS Moveit to implement Inverse Kinematics.

Code

Educational Platform for Mobile Robotics

• Inspired by the Duckietown project to develop a low cost, ROS-integrated Mobile Platform to teach Mobile Robotics.                                                                          

• Developed codebase for PID control, object detection, lane detection, obstacle avoidance etc.

• Developed Kalman filter based motion model

Gesture control of Kinova Jaco Robot Manipulator

•Implemented software interface to control the end effector position of a Kinova Jaco robot manipulator using the Leapmotion sensor.

•Used SVM's to classify hand gestures

Bomb Disposal Robot

• Designed and Developed Mobile Manipulator that can detect and defuse the live wire in bombs.

• Used the ROS Navigation stack for autonomous naviggation

Human Robot Interaction system

• ROS Node that can display various animated emotions such as happy, sad, talkative and sleepy to communicate the state of a robot.       

• Inspired by the Rethink Robotics Baxter 

About Me

I am an undergraduate researcher at the Artifical Intelligence and Mobile Robotics lab at PSG College of Technology. I am in my final year of BS Robotics and Automation Engineering.