Bhairav Mehta

I am a undergraduate senior at the University of Michigan working under Satinder Singh and Matthew Johnson-Roberson. Between January 2017 and June 2017, I left the University of Michigan to spend time at the Jet Propulsion Laboratory, working full-time under Larry H. Matthies in the Computer Vision Group. After Michigan, I will be in Montreal at MILA doing research under the supervision of Liam Paull.

My research interests are focused on the intersection of robotics and reinforcement learning, including the simulation-transfer problem, lifelong learning, and model-based reinforcement learning.

In my free time, I develop the machine learning algorithms behind an edtech startup I cofounded named Project Chronicle. We have been funded by the University of Michigan CSE Department and the 1517 Fund.

Most recently, I started writing on Medium. Like many others, I believe there is no middle-ground in learning machine learning; either you watch a high level overview, or you read a paper. Here is some of my work trying to help fix that issue: Model Based Reinforcement Learning. Similarly, here are some supplemental notes I have been writing as a TA for Michigan's Machine Learning class.

Email: bhairavm[AT]umich.edu

Github: bhairavmehta95
LinkedIn: bhairavmehta95

Active Research Projects

Multiagent Robotic RL in Gazebo
Bhairav Mehta, Matthew Johnson-Roberson
Undergraduate Senior Thesis: In Progress
Orignal Project Proposal

Reinforcement learning, and other aspects of machine learning, have allowed researchers to build robots that can function in under-modeled, dynamic, or cluttered environments. Yet, machine learning is built upon the premise that both positive and negative examples are presented to the system. In some cases, this constraint of training data is financially expensive or is dangerous to obtain. While work has been done in simulators, the bulk of work has stayed within simulators not as popular or familiar to the robotics community. This project, building upon the work of Erle Robotics aims to give researchers the ability to train teams of robots in simulation using Gazebo, a more well known physics simulator (w.r.t robotics research). The project, with a modular interface to swap out RL algorithms, robots, and environments, aims to boost interest by providing a solution that does not require porting URDF models and rebuilding robot environments.



Home Support Robotic Learning
Research under Satinder Singh
In Progress

Aiming to train robots in simulation, I have been developing agent environment code and DRL algorithms for a home support robot. I am exploring model-based algorithms for self-supervised learning.



An Information Retrieval Approach to Education
Bhairav Mehta, Adithya Ramanathan
[Paper]

We present a information retrieval approach to education and provide a end-to-end framework to go from raw text to a system where a student can learn about different topics such as History and Psychology, all while getting immediate feedback and recommendations on what to study from our system.



Past Work and Startups

ICLR 2018 Reproducability Challenge
Class Project for EECS 498: Reinforcement Learning
In Progress

Aiming to reproduce the results in "Parameter Space Noise for Exploration," my team and I entered into the ICLR 2018 Reproducability Challenge. Check out our final draft of the results here.




Perch and Stare: Autonomous Microaerial Vehicle Landing
Research under Larry H. Matthies at the JPL Computer Vision Group

I helped develop the main pipeline that transformed monocular camera images from the MAV's camera into a elevation height map. To then evaluate the usefulness and safety of landing sites, we used metrics like elevation, flatness and clutter to rank landing site candidates, and then used the quadcopter's motor controllers to autonomously land the vehicle.




Robotic Exploration of Space Team
Project Team advised by Edwin Olson
Video

My main project team at Michigan, I developed software features such as localization, teleoperation, and odometry using C++ and ROS for an autonomous mining rover, which we utilized during the Robotic Mining Competition.




Project Chronicle

There is no opportunity as big as education. It is an opportunity to make life-long learners; to excite students about the world; and to create explorers, scientists, entrepreneurs, entertainers, and engineers. But most students dread school. Many don't find relevance in their classes and many find their knowledge useless. Our mission at Project Chronicle is to empower those students with the ownership of stories and enhance their learning through the power of speech.

Project Chronicle has students record their telling of a prompted topic. Our platform analyzes the student's response and give immediate feedback on both the accuracy and delivery of the content. The application both challenges the students to understand the material at a much deeper level than required by typical homework and leaves them with stories they can easily remember and share. It gives them confidence. It gives them comprehension. It gives them a voice.



Teaching

EECS 445: Introduction to Machine Learning
Instructional Aide

Machine Learning is a great course, but there's often too much to cover, and we can't always explain everything fully when teaching the course at Michigan. As a result, I've been writing CS229-esque notes on tangential or advanced topics, for interested readers in our Introduction to Machine Learning class. Check out some work here.




Source stolen from here