Scribd
Upload
88 views3 pages

Essay

The document is Sony's full-time position application form which the applicant Chayan Rellan has filled out. In the first essay question, Chayan describes a reinforcement learning project during their summer internship where they used Q-Learning and Deep Q-Networks to solve classic control problems like cart-pole using a Raspberry Pi to control motors. In the second essay question, Chayan expresses their interest in robotics areas like path planning, position estimation and autonomous vehicles, and describes their relevant experience and skills in areas like reinforcement learning, UAV formation control and motion optimization.

Uploaded by

Chayan Rellan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
88 views3 pages

Essay

The document is Sony's full-time position application form which the applicant Chayan Rellan has filled out. In the first essay question, Chayan describes a reinforcement learning project during their summer internship where they used Q-Learning and Deep Q-Networks to solve classic control problems like cart-pole using a Raspberry Pi to control motors. In the second essay question, Chayan expresses their interest in robotics areas like path planning, position estimation and autonomous vehicles, and describes their relevant experience and skills in areas like reinforcement learning, UAV formation control and motion optimization.

Uploaded by

Chayan Rellan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Sony Full-time Position Application Form (Your own resume must be attached)

↓↓ Must be filled by the Applicant

Position Number
Priority 1 AUTONOMOUS MOBILE SYSTEM

Priority 2 MACHINE LEARNING TECHNOLOGIES FOR ROBOT CONTROL

Priority 3 MACHINE LEARNING ENGINEER

General Information
Name CHAYAN RELLAN

Essay Questions

1. Please describe one project or research that you focused on. Must include role you played, difficulties, learnings and
reflection. (About 800 words)
My project this summer was to solve the cart pole inverted pendulum model using
reinforcement learning algorithm using the open AI Gym interface. The summer
internship at New York University (NYU) has introduced me to the advancements in the
field of robotics throughout the world. This facilitated me to realize international
experience in conjunction with the data and ability that I have learnt and developed
throughout the course of the internship. Moreover, it enabled me to acquaint myself with
the latest technology and helped me to interact with the professionals from all around the
globe.
The inverted pendulum swing-up problem is one of the most famous and classic problems
in control theory. The main objective of this project was to balance an inverted pendulum
mounted on a cart by moving the cart horizontally using reinforcement learning.
Reinforcement Learning is of two types – model based and model free. A model free
approach was used while solving the above problem where the agent explores the
environment and learns the optimal solution to the problem unlike the model based RL
where a model has to be designed in advance.
My first task was to create an environment consisting of a 3 x 3 cell with a robot at the
starting position and a cell where the treasure was placed. My task was to find the best
solution for picking up the treasure and putting it back at the starting point. This task was
accomplished through the use of Q Learning (a free model of reinforcement learning)
implemented in Python. The second challenge was to learn Deep Reinforcement learning
(a mixture of Q Learning and Neural Networks) and use it with the assistance of the Open
AI gym interface to solve the Cart Pole control problem.
Thirdly, I was asked to use serial-socket connection (TTL using sockets) to drive the
motor driver using Raspberry Pi. This was one of the key steps because we needed to
change the motor velocity and direction depending on the RL algorithm output. Nothing
about governing the motor using raspberry pi is mentioned in the documentation. Instead,
they've created a library for Arduino that makes it simpler to use the Arduino to control
the motor. Since we want to control the motor based on the algorithm's results, we had to
pair it with raspberry pi. It took a lot of time because no one has yet worked with the
Polulu Qik motor driver using raspberry pi and hence we had to subjugate almost every
source available online as well as post queries on forums.
Finally, I was allocated the job of constructing the cartpole mechanism's practical model.
This comprises of two rails installed with a coupler on a 3D printed platform to which the
pendulum is suspended. This coupler is linked to the motor powered by the motor driver.
A camera would also be put to record the cart pole states that act as an input to the RL
algorithm.
My contributions to the RL Inverted Pendulum project include the DQN algorithm, TTL
serial-socket communication which involves governing the motor using raspberry pi from
another system and the 3D hardware construction which involves a chain drive
mechanism for balancing the pendulum as mentioned above.
2. Please describe what you would like to contribute through the position you are applying in Sony, from the aspect of
knowledge/skills or experience. (About 500 words)

I was quite impelled by the job descriptions at SONY, especially in the field of
robotics and was enthralled by a few of them. I am interested in the field of
Robotics especially in path planning, position estimation and autonomous
vehicles. I feel this would be a great opportunity for me learn, develop and
implement the desired skills.
As understood by me, the motif for research in this particular area is to develop a
vision-based robotic system that allows a mobile robot to avoid collision with
dynamic obstacles i.e. obstacles whose position changes with time in space. I
would like to work on optimizing the algorithms that are being used currently as
well as develop better algorithms for faster and error-free reflexes.
I would like to share with you that I am currently working on semester project
which deals with motion of a UAV between two points along the least potential
path in order to optimize the battery usage. Collision avoidance also plays a major
role in such situations. I have previously worked at New York University on
Reinforcement Learning under professor Vikram Kapila and Indian Institute of
Science, Bengaluru in the Mobile Robotics Lab with professor Debasish Ghose,
where I worked on the feeback formation control algorithm for UAV Swarms.
Working at SONY would introduce me to the advancements in the field of
analysis throughout the world. It will enable me to acquaint myself with the latest
technology and would help me to interact with the professionals from all around
the globe.

You might also like