My research projects are often fuellled by my own curiousity and passion of robots.
The tactile capabilities of humans are so advanced that we can literally grasp objects without using our eyes. Consider grabbing a keychain or a coin from your pocket. We never look into our pockets, yet we can grab them everytime. This work explores how to enable robots to do the same. Read more
This work explores the developing new reinforcement learning policies to tackle the problems which are generally solved using the traditional model based control methods.
The above video shows a kuka robot trying to reach the position of the green ball. The environment accepts joint angles as input and returns a reward depending on the distance between the end -effector and the target green ball. Since the input space is continuous, a DDPG agent is used for training.
This research aims to mimic the reflex action of cats, which enables them to land safely on all the four limbs. A 2DOF active spine enables the quadruped to rotate the frontal and dorsal parts of the torso. Making usage of the conservation of momentum, the quadruped can orient itself in mid air. As of now, the concept is verified in simulation and the prototyping is being done. Read more
This project explored the design, simulation and prototyping of a linear clutch which controls force transmitted from the actuator to the robot links. The force was coupled through MR fluid and the level of magnetization controlled the amount of force transmitted.
This is the first research project I took up and completed in a very short of time, yet accepted for a top tier conference. You can read how I managed to pull this off and what all i studied by working alone in my blog
Robot Arm for Testing of Touchscreen Applications Pub. No.: WO/2017/051263 International Application No.: PCT/IB2016/05329