Domain Interests
- Robot Motion Planning
- Multiple Mobile Manipulator Systems
- Trajectory Planning
Research
My research intersts are on designing motion planning algorithms for multi-robot system that can deal with advance problems like cooperative manipulation, junction congestion, collaborative excavation, etc. I am inclined to enable robots or team of robots to perform to tasks that are hazardous and impossible for humans to perform.
It is difficult for humans to work in underwater environments due to restricted reachability and unfavourable conditions. Performing construction activities is even more challenging as human to machine collaboration becomes more difficult. Hence, to facilitate cooperative transportation of complex structures we have developed a novel cooperative path planning algorithm: Intersection of Manifolds RRT* (IMaRRT*).
We are keen to extend this work to incorporate environment dynamics.
Research Advisor
Featured Projects
-
Pick Me Up
As a stepping stone to full blown robots treasure hunt, we programed the Fetch Robot to detect a specific object, localize it, navigate to reach a safe distance from the object and successfully pick it up. After assembling a physical setup, a map was generated and fetch robot was localized in the environment. Carton boxes were used as markers for localizing the robot. Using computer vision techniques in openCV a mask was generated to extract the object. Erosion and dilation were used to get object contour. Using MoveIt motion planner successful robot path was generated. RRT planner was used to generate the path for robotic arm to pick up the object.
-
Open the Door
As a part of setting up a robotic arm testbed infrastructure, I simulated a Kinova Arm in Gazebo to open a door by integrating moveIt!, Rviz and Gazebo with ROS. Python was used as the major programming language. One of my interesting learning in this project was the process of creating a xacro from an stl and import it as an object.
-
Train me Up
External cameras typically require an offline calibration process that is notoriously tedious, sensitive, and error-prone. A plausible solution to this is training a neural network that can successfully detect robot pose in camera frame. I developed a pipeline in Unreal Engine to generate synthetic data for a Kinova Arm. This data was further used to train a neural network that can predict Kinova arm pose.
-
Portable Treatment Device
Prototyped proof of concept for miniaturized imaging technique that can fit in a unit square meter. This work can be easily extended to treat people suffering from psoriasis by incorporation. of UVC LED panel. From designing the mechanical system assembly I extended my work by architecting the embedded hardware around Raspberry Pi. Four cameras were multiplexed to a single raspberry pi. I also developed an algorithm for stitching together images using SIFT feature alignment techniques. I used openCV library while developing this project.