smarpall.github.io

Portfolio


Projects

Multi-Robot Decentralized Localization

![MRDGE](/smarpall.github.io/images/mrdge.gif) </center>

Internship work at IIT Bombay: Simulating Multi-Robot Decentralized Graph Exploration on ROS-Gazebo environment. Robots Cooperate to completely explore an environment with decentralized control and indirect communication using landmark beacons. This project was completed under the guidance of Dr.Leena Vacchani and Dr.Arpita Sinha at IIT Bombay, Systems and Control Engineering Department.


Deep Recurrent Neural Network based motion planning


Internship at IIT Kanpur: Mimicking RRT Motion planner using Deep Recurrent Neural Networks as well as learning inverse kinematics using Deep Feed Forward Networks. Data was collected from the UR5 simulator on Gazebo and used to train a Deep Recurrent Neural Network for Motion Planning as well as to learn the inverse kinematics function for the UR5 robot arm. This project was completed under the guidance of Dr. Laxmidhar Behera at IIT Kanpur, Electrical Engineering Department


Navigation using DQN
Deep Q Networks were used to train an agent using reinforcement learning to learn to pick-up reward objects in the environment. This project was completed within the Deep Reinforcement Learning Nanodegree, Udacity.


Simulation of Search algorithms

Simulation of Search
CSE 571( Artificial Intelligence) Project This project simulates a battery deficient robot which uses search algorithms to solve a maze-like environment. The project uses the ROS-Gazebo platform


Deep Deterministic Policy gradient Implementation

DDPG
Deep Deterministic Policy Gradient algorithm was used to train an agent using reinforcement learning to solve a robot-reach task. This project was completed within the Deep Reinforcement Learning Nanodegree, Udacity.