R3plica

Meet Khush Agrawal

Robotics | Reinforcement Learning @ IVLABS

Nagpur, Maharashtra, India

Khush Agrawal interest lies in Reinforcement Learning, particularly in its application to Robotics.

He had worked on Machine Learning for a while now, and have developed an ardent interest in Reinforcement Learning by working on multiple robotics-related projects.

His wish to contribute to research in Robotics and Reinforcement Learning.

Here the portfolio/website where all his projects can be found: Khush: Roboticist

What is your goal with your studies in robotics?

I aim at becoming a full-stack roboticist. I started this journey right from my freshman year, where I scratched the surface of electronic-circuit designing and programming microcontrollers.

By the end of my freshmen year, I developed a robot from scratch capable of localizing itself.

To learn about how I can make a robot perceive the world around itself, and act smartly, I was motivated to study Machine Learning.

By the end of my sophomore year, I developed an autonomous stair climbing robot and a person following robot.

Moving up the stack, I always felt that self-study was an unstructured way of learning. Taking up formal courses on topics like Machine Learning and robotics could potentially create a given structure to my learning process.

I plan to investigate the field of Reinforcement Learning by pursuing a master's degree.

Describe the skills you think are needed to be a roboticist?

Robotics is broadly a blend of 3 fields, namely Mechanical Engineering (The Body), Electrical Engineering (The Nervous System), and Computer Science (The Brain).

To be a full-stack roboticist, one needs to master these fields.

Mechanical engineers are responsible for robotic manipulations. This is done by computing the joint parameters and planning trajectories that achieve a specified position of the end-effector.

Electrical engineering plays a role in designing the overall circuitry required to connect all the sensors and actuators to the brain of the robot. Also, a good understanding of designing controllers comes under the same branch.

Computer Science is the backend of a robotic system, responsible for decision making. All the generated sensory data is processed to produce some meaningful actions. This processing can be done through various approaches, which include Computer Vision, Artificial Intelligence, or traditional programming paradigms.

Besides this, a good understanding of mathematics is instrumental in pursuing a quality research carrier. One needs to master these skills to be a roboticist.

Mention some areas that robotics can be applied?

Robotics finds its applications in a large number of areas. These included human assistants, companions, study buddies, etc.

Also, they can be deployed to areas that are danger prone, including operations in factories, an inspection of heat wells, furnaces.

Inspection Robotics is an exponentially growing field.

Several industries have incorporated robots for inspection. Robots allow safe and faster operation in places humans could never go. With cognitive and physical capabilities, one could never imagine.

Robots also find their application in the area of defense. Using robots could save hundreds of lives. Apart from this, robots are incredibly efficient warehouse managing resources.

Amazon has recently automated its warehouse using a swarm of robots for management.

Given the recent advances in robotic arms, there have also been multiple demonstrations of using robots in performing complex medical surgeries.

How can robots help fight the spread of a deadly outbreak?

Let us consider the current outbreak of coronavirus.

Robots can be used in numerous ways to avoid this threat to human life.

To prevent the spread of coronavirus (and everything else) through hospitals, keeping surfaces disinfected is incredibly important.

It is an ideal task for autonomous robots.

Besides this, robots can be used to supply food to the quarantined people, other basic requirements to isolated people.

This can significantly serve the purpose of social distancing and avoid the spread of the virus.

What role models have been your inspiration to build robots?

I am deeply inspired by Prof. Christopher Atkeson's (faculty at Robotics Institute, Carnegie Mellon University) research, particularly the BayMax Robot.

It has also been featured in a Disney Movie: Big Hero 6. Baymax is a companion robot that also takes personal and health care.

Such robots can be used to raise our standard of living. It can help dress, people, feed them and comb their hair, and many more.

The motivation behind the development of this robot (BayMax) is inspiring me, and driving me towards my goal of becoming a full-stack roboticist and contributing to society.

There are any Robotics projects that want to share your experience with us?

Autonomous Stair Climbing Robot: In my freshmen year, I worked on a differential drive robot, capable of localizing itself in a two-dimensional plane.

However, I realized that the differential drive robot I fabricated had the limitation of being able to transverse only on uniform terrain.

To overcome this limitation, I started to work on an All-Terrain Ground Vehicle.

Also, I took up the challenge to make it autonomously climb a stair, as it is known to be a problem with immense significance in indoor navigation.

I developed a two-stage segmentation and behavioral cloning model pipeline to accomplish the task.

I faced challenges like generating data for the behavioral cloning model, selecting the input features, where I used the domain knowledge, and heuristically generated empirical results to solve the problem.

This work has been published in the International Journal of Semantic Computing Vol. 13, No. 4 (2019) 1–16.

Person Following Robot: In my sophomore year, I aimed at deploying robots to solve a societal issue, and I started to work on a person following robot. It has several applications like assisting the elderly, surveillance applications, etc.

Using Tracking and Detection Algorithms, I developed a robot to follow a person in a dynamic environment.

The challenges incurred during the development included occlusion handling, where I used a depth camera to detect and handle occlusion apart from estimating the target coordinates.

This research has been presented in the International Conference on Advances in Mechanical Engineering 2020 and will be published by Springer.

Here the demonstration: Person Following Mobile Robot