BITS R&D caught up with Rishikesh Vanarse, a CSE student from BITS Goa (2017 Batch), who remotely completed his thesis in Visual-Inertial SLAM for Autonomous Underwater Robots at University of Nevada, Reno- and later NTNU, Norway
How were you introduced to Robotics? What helped you narrow down to a specific topic to pursue research in?
I’d been an active member of the Electronics & Robotics Club of BITS Goa (ERC) since the beginning of my first year. I worked on several projects as a part of this club, where I got a chance to design and automate a variety of robots. I also got an opportunity to learn a lot about the field through seniors and through my own experience. Although I knew that I was definitely interested in robotics, I had also taken some time out of my work in ERC to explore Deep Learning and Computer Vision. By the end of my second year, I was sure that I wanted to pursue a career in robotics research.
Deciding a more narrow area within robotics was not very difficult at this point. As my discipline was Computer Science, I wanted to keep the thesis relevant to the automation side of robotics. I had experience in working on autonomous navigation and I found the topic very interesting. I had also worked on computer vision in other course-projects, smaller club-projects and my PS1. These factors led me to shortlist the following areas:
- Computer vision in robotics
- Path Planning for autonomous robots
Once I had an idea on what areas I wish to work on, I still had a few months to spare before I start applying. I spent these months taking up as many projects as I could that were directly relevant to what I’m applying for. The idea here was to strengthen my profile in these areas before I start mailing.
Once you decided your field, how did find your desired laboratories to work in? How did you apply to these places?
I began mailing around October/November 2019 for an August 2020 thesis. I started with moderately sought-after universities and listed the professors that work in a robotics lab or research group. Each professor has a website or a university profile which lists their interests. I also visited their Google Scholar profiles to check if they have any recently published papers that align with my interests. I shortlisted the ones who were interested in the areas I had chosen, are currently working in those areas and have an h-index of around 25-50. Over time, I kept appending to this list even after starting the actual mailing.
I followed a highly personalized mailing approach for the applications. I spent around a week per professor, learning about their current research through their papers. Finding a recent preprint is always immensely useful to know exactly what to apply for. The ‘future work’ section gives a clear idea about what the lab or professor plans to work on in the coming months. I also connected with grad students from each lab on LinkedIn/Gmail. By talking to them, I tried to find out where they may require students and the kind of work interns would do. Finally, I spoke to seniors and others from my social circles who had worked with some of these labs, to get useful pointers on what to say in my mail. Once I had all this information, I sent emails that mentioned what I know of the prof’s work, how it aligns with my interests/experience and how I plan to contribute to it.
What finally led you to join Dr. Kostas Alexis’s Autonomous Robotics Lab?
An ERC senior, who I had worked with closely, was pursuing his thesis in the Autonomous Robotics Lab of University of Nevada, Reno. Through him, I learnt that their lab is working on Perception for Aerial Robots. Another BITS alumnus who I had worked with previously was a PhD student in the same lab. I was recommended to the professor by these seniors, which gave me an edge. I had my interview in December and got my offer letter at the beginning of 2020. The professor (Dr. Kostas Alexis) was well-known in the robotics community and was doing very interesting work on autonomous navigation in challenging GPS-denied environments. Since this was exactly what I wanted, I further decided to graduate early and extend the thesis to a yearlong one.
Tell us what you worked on in the lab!
The first project that I worked on at ARL was in agriculture robotics. This was a small month-long project. Here, I worked with drone-images taken with a ‘multi-spectral’ camera that records images in different infrared and visual spectra. My work involved using principles from satellite & medical imagery to combine multispectral images of a field taken from different angles into a single map that gives highly precise information about the crop. This was interesting because IR images look very different from visual images, making it tricky to determine correspondences between two images from different modalities.
My second project was on navigation for autonomous underwater robots. This is the project that I wrote my thesis on. The project was related to ‘Simultaneous Localization And Mapping’ (SLAM). This is a process where a robot tries to build a map of its surroundings while trying to estimate its own location at the same time. Computer vision is often used for this. However, the nature of light transmission in water degrades the visual inputs to the robot, making SLAM challenging. In my project, I developed a method that uses stereo-images and the properties of underwater light propagation to reverse the degradations in underwater images. I then incorporated this image-enhancement in existing visual-inertial SLAM methods to improve trajectory estimation and reduce reliance on sensors like the IMU (motion sensor).
What were some things you learnt there about the method of doing research?
When I started working with ARL, I realized that the nature and the methodology of working in such labs is very different from the way we had been working on low-level student-research projects back in BITS. Before even deciding what question the research will try to answer, a significant amount of effort is put into thorough literature reviews and testing things out. Tasks such as deciding what questions to ask, deciding what tests to conduct, how to verify something, etc. were carried out far more systematically than I had seen before. Similarly, rigorous experiments were performed before reaching even seemingly simple conclusions. I noticed an overall mindset of being absolutely sure about the current outcome before proceeding further (which is something that I noticed in my later research work outside ARL too). Another important fact that I learnt was that in most research, you never know whether you will achieve a favorable result and whether something will work, until you actually try it. Knowing that something doesn’t work is also progress, and sometimes, reaching this point may take months too.
The other key difference that I saw was that funding received by research labs in the US, especially in fields like robotics, is much higher than that received by most labs in India. As a result, the research questions that my lab worked on were very different from those I had seen in India. Robots, sensors, etc. relevant to both my projects were far more expensive than what I had used before, and of a much higher quality. Therefore, fewer things were “out of scope”.
The researchers at the lab were all highly proficient, and had in-depth theoretical knowledge as well practical skill in the field. My supervisor, mentors and colleagues were all very helpful and it was a great experience working with them.
How did the COVID-19 pandemic affect your thesis?
This is the part of this interview that I hope becomes irrelevant soon. Being from the ill-fated 2017-2021 batch, we had never heard of COVID when we got our offers and were deep into the pandemic by the time our theses began. Neither students nor professors had much of an idea on how to adapt to the new circumstances and this affected the thesis for most of us. If the pandemic is still here while you read this, here’s how it can affect your thesis:
The erratic nature of COVID waves, changing travel restrictions, etc. lead to several uncertainties and thesis plans become less concrete. Most hurdles arise from having to work remotely. Conducting experiments becomes difficult and access to necessary facilities/equipment becomes negligible. In many fields, remote work may not even be possible, which may lead to drastic changes in the research topic, or in worse cases, cancellation of the thesis itself. Although my thesis topic changed before I joined, I was lucky to get a topic that I was equally interested in. However, this was not very common among many others in the batch. Further, communication and doubt-solving can become much more difficult, especially when you are in the opposite time zone. This makes it a bit harder to understand what is expected of you as well as to convey your own thought process. If you have to work alone, it increases the chances of frequently going in tangential directions, and then spending time reworking. Lastly, your stipends may be scrubbed, since universities usually pay the stipends primarily to cover your cost of living.
Despite the complications, the right thesis will still be a great learning experience. My supervisor and mentors took substantial efforts to ensure that the work gets carried out smoothly even as a remote thesis. Thus, though remote, it was some of the most interesting work that I have done so far.
What are your plans for the future? Where will we see your work next?
I plan to continue working in robotics research. I am currently working on IMU-based navigation for autonomous vehicles with the Autonomous Systems Lab of UC Santa Cruz as a GSoC project. I am also working on some of my own research in Robot Swarms. In the coming months, I plan to work in robotics R&D as a job or as an RA in a lab (Author’s Note: Rishikesh is now working as an RA in the BioRobotics Lab at Carnegie Mellon University!), before applying to PhD positions for Fall 2022.
All the best, Rishikesh!
BITS R&D 
Leave a comment