I am current working on dynamic manipulation strategies. More on this later.
Large-Scale Optimization: Mixed-Palletizing and Logistics Operations
Palletizing is a very important problem in today’s manufacturing industries. It is considered a NP-hard problem. Packaging goods is almost a part of every industry and efficient packaging can reduce costs, transportation costs.
Mixed palletizing is a hard problem and benchmarking heuristics is very important, so that a search algorithm can be designed to place boxes in a pallet. We plan to run a demo with the KUKA KR5 at ICRA 2010.
Modeling 3D data and grasp planning
The MESA 3D range sensor is an excellent sensor for getting a 3D point cloud. To interact with objects we need to detect high level object features like planes, shape. We are currently trying to fit the 3D range sensor data to planes, which can then be interacted with.
We want to learn how robots can interact with new objects based on previous known models. We are trying to use a Bayesian learning algorithm to compare and grasp.
Dynamics of a balancing humanoid robot and its applications
Humanoid robots have complicated dynamics. I was working on a balancing robot manipulator. What makes this robot interesting to study is that, it closely resembles dynamics of a humanoid robot and suffers less from an actual humanoid because of its less complicated mechanics. Sparky is a two-wheeled balancing robot manipulator. I studied how it can be used to perform manipulation tasks using pushing and pulling. We found that such a robot is very effective and can perform almost human like tasks by pushing objects heavier than itself.
A Communication Model for Multi-robot systems
Single robots are insufficient to perform multiple or complex tasks such as exploration or de-mining, multiple robots can be used to increase efficiency and reliability. Here coordination and planning among the robots is important. The project focuses on such an issue. There are many researchers who have proposed different models for multi-robot coordination. However, there are only a few which will dynamically handle faulty sensor issues or loss of robot capability due to faulty hardware. We have tried to implement such a model so that we can realize the power of the multi-robot system along with the difficulties involved in a multi-robot system.
I worked for the Robocup 2008 in BORG Lab at Georgia Tech. This is the standard platform league that will have the Nao bots from Aldebran Robotics.
I worked on some basic computer vision algorithms and the communication framework. I am currently developing a localization algorithm based on particle filters and multirobot interaction.
We stood second at the Robocup 2008 competition in China. Some of our videos can be seen on Youtube.
I am currently working on a mobile manipulator platform. The mobile platform is a Segway RMP and the manipulator is a KUKA LWM (Light Weight Manipulator). Our task for the mobile manipulator is to grab a CD from a user and play it in a nearby CD player, and then return the CD back to the user.
The main problems involved in this project are arm motion control using visual servoing and force control. Also, the Segway is a two-wheeled platform and hence it does not remain stable when the arm moves. We call it the inverted pendulum problem, where we try to find the position of the tool of the arm by using various techniques.