Rohan Kamath

In this project, I led a team of 4 cross discipline engineers to design, fabricate and program a 3-DoF robot arm to pick and place chess pieces.
We were responsible for the complete mechanical design as well as the programming of the robot. We used standard plug and play electronics as hardware interfaces.
The robot was an RRP manipulator. The first joint was controlled by a DC motor via cable drive. The second joint used a stepper motor with cable drive. The third joint used a DC motor… Show more

In this project, I led a team of 4 cross discipline engineers to design, fabricate and program a 3-DoF robot arm to pick and place chess pieces.
We were responsible for the complete mechanical design as well as the programming of the robot. We used standard plug and play electronics as hardware interfaces.
The robot was an RRP manipulator. The first joint was controlled by a DC motor via cable drive. The second joint used a stepper motor with cable drive. The third joint used a DC motor that used a cable to raise the end effector by winding it and lowered it via gravity.
A discretized version of PID control was used for the DC motors. Using this customized algorithm, we were able to attain 99.5%+ precision almost every single time. The stepper motor was controlled at varying step rates and velocities.
We computed joint angles via inverse kinematics for all the joints.
The end-effector was a custom gripper that was controlled by a servo motor and a pair or spur gears for reversing direction.
The robot was initially homed to a start position from which all the IK had been calculated. It was then given pick and place co-ordinates of the chess board(eg. [8,8] to [4,4]) via a simple GUI.
Other projects done in this class involved designing and prototyping various gears trains, friction differentials, belt drives, cable drives, remore center of motion, rigid linkages, etc. Show less

I worked with Prof. Oussama Khatib in the Artificial Intelligence Lab at Stanford University on building a robotic system to autonomously etch and carve shapes into blocks of wax. This project was part of the Experimental Robotics class.
In this project we used a combination of Joint Space and Task space trajectories on the PUMA robot arm to perform the dual task of etching and carving.
I designed an end effector that could carry a variety of tools. After some experimentation, we… Show more

I worked with Prof. Oussama Khatib in the Artificial Intelligence Lab at Stanford University on building a robotic system to autonomously etch and carve shapes into blocks of wax. This project was part of the Experimental Robotics class.
In this project we used a combination of Joint Space and Task space trajectories on the PUMA robot arm to perform the dual task of etching and carving.
I designed an end effector that could carry a variety of tools. After some experimentation, we finalized on a tool that could perform both tasks with only a change in orientation.
An input file containing the lines to be etched and the discrete points to be carved were given as input to the algorithm.
The robot could then be calibrated and it would carve the given pattern accordingly.
The main challenge in this project was the 3D work space and ensuring the end effector was always in contact with the wax. Show less

I worked with Prof. Oussama Khatib in the Artificial Intelligence Lab at Stanford University on building a robotic system to autonomously clean up a dinner table. This project was also part of the Machine Learning class with Prof. Andrew Ng.
In this project we use a Microsoft Kinect sensor mounted on top of the PR2 robot (Willow Garage Inc.) to obtain RGB and depth images of table top objects.
K-means clustering is used to cluster the objects and a multi class SVM classifies the objects.… Show more

I worked with Prof. Oussama Khatib in the Artificial Intelligence Lab at Stanford University on building a robotic system to autonomously clean up a dinner table. This project was also part of the Machine Learning class with Prof. Andrew Ng.
In this project we use a Microsoft Kinect sensor mounted on top of the PR2 robot (Willow Garage Inc.) to obtain RGB and depth images of table top objects.
K-means clustering is used to cluster the objects and a multi class SVM classifies the objects. 500 images of cups, bowls ad plates were used to train the SVM using a simplified version of shape context features.
Further, the algorithm attempts to recognize these general dining table objects and locate their positions.
Finally, the robot picks up the objects and place them in separate boxes based on their type. Show less