My long term research goals are towards developing robust perception and interaction models for robots. My background in Mechanical Engineering, especially design, drives me to work on designing Human-Robot interaction interfaces to aid in development of versatile robots that can work alongside humans in unstructured environments. Currently, I am working on the design and control of a flapping wing Micro Aerial Vehicle.
We humans have always dreamed of having robots work alongside assisting in tasks but such high level robots have been limited
to science fiction for example Data from Star trek. Most Recent progress in that direction can be seen in humanoid SCHAFT
by researchers from the University of Tokyo. SCHAFT robot showed robustness in control and also advanced planning and
perception. Robot perception in particular is quite challenging because of the constant stream of data from various sensors
which the robot needs to understand. With the advances in computing technologies, we now have the required tools to process
that data and help the robot understand its surroundings. This can have a wide range of applications from minimally invasive
surgeries to Human Assistance robots that interact in human environments. I believe Assistive robots can help make lives
of elders and disabled easier. I want to tackle this problem from perception and control point of view.
Dato’s GraphLab Create was used to train a Deep Neural Network classifier to recognize and classify 7 different emotions. A facial key-points database was compiled for all 213 images and these features were then used to classify emotions. Accuracy increased from 52.7% to 74%. work in progress.
As part of the final year Project work requirement, my team worked on building a flapping wing micro aerial vehicle. My
significant contribution to this project was in the form of designing and synthesis of a flapping mechanism. I also worked
on designing a testbed for the mav to experimentally determine the lift generated and draw parallels between theoretical
and experimental data. I am currently working towards miniaturization of the mechanism and developing a passive control
A simple lattice Boltzmann implementation has been created to demonstrate the various applications of Lattice Boltzmann
method. The history of LB method is discussed briefly, its derivation from the Lattice gas automata and the collision
model BhatnagarGrossKrook (BGK). Further, various types of boundary conditions are discussed and finally an implementation
of Lattice Boltzmann Method using simple bounce back boundary conditions was used to simulate flow over a cylinder.
code (coming soon.)
To provide certain degree of autonomy to the MAV and aid in obstacle avoidance, I worked on developing an Optical flow
algorithm. An optical flow algorithm works by trying to calculate the rate of motion of a feature between frames there
by estimating proximity of the object and time to collision so that the MAV can perform necessary maneuvers to avoid
collision. This particular implementation uses LucasKanade method to determine optical flow and is implemented in MATLAB.
code (coming soon)
Aim was to build a gesture controlled robotic arm with 4 + 5 degrees of freedom. The robotic arm has 4 links that help in the motion of the arm to a given point in 3D space and the anthropomorphic end effector consisting of 5 links resembling a human hand are used to hold objects. Servo motors are used as controlling elements for providing the different degrees of freedom to the robot arm. The servo motors are controlled by an Arduino micro-controller.
The Robotic arm is controlled by 3 high-torque servo motors and the end effector is controlled by 5 low-torque servo motors. The arduino controls the servos based on the signals obtained from a controller glove. The controller glove consists of 2 force sensors to sense the position of the fingers and an accelerometer to sense the angle of the arm.
The Robot arm moves according to the position of the human hand and the end effector can be used to manipulate objects in 3D space. Robotic arm is useful for industrial applications like welding, painting, assembly, pick and place, product inspection and testing with great accuracy. The project is very useful in gaining new experience and knowledge on robot arm fabrication and programming.
I participated in the Artificial Intelligence MOOC offered by UC Berkeley and worked on implementing various search algorithms, Constraint Satisfaction problems. I gained insight into the working of Markov Decision processes (MDPs), Hidden Markov Models (HMM) and reinforcement learning.
After participating in Machine Learning MOOC offered by Stanford (Prof. Andrew NG), I worked on implementing a machine learning model to perform OCR on images.
I have also worked on various small projects on arduino programming, web scraping, object recognition and tracking, bot building - line follower and obstacle avoider bots and various others related to Robotics, Computer Vision and machine learning.