Wired to succeed

0
567
Rajat Talak was recently awarded a PhD from Massachusetts Institute of Technology (MIT), USA. In conversation with NT KURIOCITY, he speaks about his thesis, the new field of networked autonomy, and more

ANNA FERNANDES | NT KURIOCITY

In a proud moment for the state, Margao’s Rajat Talak earned a PhD for his thesis on ‘Information Exchange and Robust Learning Algorithms for Networked Autonomy’ from Massachusetts Institute of Technology (MIT), USA.

Armed with a BTech from the Department of Electronics and Communication Engineering of the National Institute of Technology Karnataka (NITK Surathkal) and an MSc in Electrical Communication Engineering from the Indian Institute of Science (IISc) Bengaluru, Talak, since joining MIT in 2014, has been working in the new field of networked autonomy.

Along with the publication of several papers, he has also co-authored a book titled ‘Age of Information: A New Metric for Information Freshness’ with his advisor Eytan Modiano and colleagues Yin Sun and Igor Kadota, published by Morgan and Claypool Publishers.

Excerpts from the interview

Q. What inspired you to venture into this field?

I am an engineer by qualification who likes to work on the mathematics of a given engineering problem. I was inspired into this direction during my undergraduate years at NITK Surathkal where I realised that a major fundamental bottleneck in designing an engineering system is the difficulty in solving the mathematics associated with it.

To give an example, consider designing a simple mechanical system – say a pendulum clock – without knowing Newton’s laws! It would be an arduous task of trial and error – not to mention the money wasted in building repeated prototypes. It is Newton’s laws that help us express the workings of the mechanical system as a bunch of mathematical equations, solving which we can design the system on paper. This principle holds for most engineering systems today – from designing computers and algorithms that run them to designing planes and autonomous cars. Expressing their functionalities in terms of mathematics, finding out key patterns in the expressed mathematical equations, and solving them makes engineering systems that would have been previously inconceivable, practically possible. 

Today’s leap in artificial intelligence, the fact that you can find the shortest route from your home to your office on an app, the fact that your phone can recognise your face with its camera, that we can talk without much interruption and almost at a free of cost – from Boston, US, to Goa – is made possible because a lot of mathematical challenges that were ensued in the engineering problems have been solved. 

Having said that, every aspiration to build a new type of machine, or a new kind of computer, or a new algorithm that enables a previously considered impossible-to-accomplish task, will throw up new mathematical equations and challenges to solve.

Q. In a nutshell, how would you explain the essence of your thesis to a layperson?

At MIT, during my PhD, I worked on problems related to networked autonomy. Networked autonomy deals with problems in which several autonomous agents interact via a communication network to cooperatively perform intelligent tasks. Intelligent tasks include joint motion planning – an algorithm that tells the robots where to go next, collision avoidance – an algorithm that averts collisions, building a 3D map of the environment by exchanging camera data, and many more – basically, anything ‘intelligent’ that you can think of, that humans effortlessly do, which is also useful.

Over the last decade, we have seen the advent of self-driving cars; civilian and personal drones – that are thought to be used in city policing, Amazon/Flipkart home deliveries, or just for filming; industrial robots – that have replaced human labour due to reduced cost and increased accuracy and efficiency; and many more. On the other hand, wireless communication chipsets are now available at a very low cost. The two together imply that we are heading towards a future where several autonomous agents or robots are likely to interact via a communication network to cooperatively perform intelligent tasks.

It is the aspiration now to build such systems that has thrown up a lot of new challenges – in the mathematics of it. My work relied on formulating and solving them.

Two specific challenges I tried to address were – firstly, how should the autonomous agents or robots exchange information? The requirements for information exchange between robots are quite different from that of information exchange designed to receive e-mails or make calls over Zoom and Skype.

And secondly, how should the robots learn, given that what they learn need to be robust to unforeseen errors. Safety guarantees are critical to the legal approval of technologies such as self-driving cars. We proposed a new theory of uncertainty variables, which could be applied to develop learning algorithms that provide such guarantees.

Q. What would you say is the future of communication networks, autonomy, and networked autonomy?

Over the last two decades, we have seen an exponential rise in the use of communication networks. In the early 2000s, I remember my grandfather had a very bulky looking cell phone and we all had dialable red-coloured telephones. Today, in the pandemic, schools and colleges are conducting lessons online and youngsters in schools probably wonder why we use the word ‘dial’. The future in communication networks are a bunch of technologies, including 5G, that will enable even higher speeds and very low delay in connectivity. This would mean that we would be able to stream an HD quality video when we travel on the bus or we may be able to watch a cricket match in augmented reality.

The autonomous systems are just starting to appear in the consumer market. However, for some time now, the manufacturing industry has moved towards automating its production line. Increasingly, you will see autonomous bots that will clean your house, automated restaurants that will take your order, prepare food, and serve them to you at your table, and in a decade from now self-driving cars! While the first two are already in existence, most experts in the field estimate that self-driving cars will be inevitable in about 10-15 years. Networked autonomy will be a requirement to enable many of these technologies. For example, self-driving cars will have to prevent road accidents, make a decision on whether to overtake the truck in front or not, and crossroad intersections. All of this will have to be accomplished by the self-driving cars talking to one another – without human intervention.

Q. Do you see an expansion in their applications in fields such as healthcare, which are emerging as vital sectors today? 

Absolutely. One example that I can think of is telesurgery. Robotic systems being more precise – and, unlike humans, can remain so even after hours of work – have been conceived to perform delicate procedures in surgery. With low latency, communication enabled by an expert in another city or country will be able to perform such procedures remotely – making use of the robotic system (that performs the surgery) and a low latency communication network.

Q. What advice would you give Goans

interested in a career in research?

A career in research is a career of lifelong learning and exploration. If you are serious about such a career I would strongly recommend that you inculcate certain life-skills that will prepare you innately for such a path. Develop a reading habit – read good books that challenge you, in your thinking. Develop a habit for rigour and perfection. Nothing you do should be less than the best effort. Look for important problems around you and attempt to formulate a solution for them – every time in your mind. Know that no age is too old to learn new things, gain new expertise, or acquire new skills. Use technologies around you (such as the internet) to explore and learn. Search for good mentors with whom you can learn valuable lessons of the profession. And the most difficult thing of all – be extremely self-critical. I have seen that Goans have a special weak spot when it comes to this, but it is not something that is impossible to overcome.

Q. What’s next for you?

I am joining as a post-doctoral researcher at SPARKlab run by professor Luca Carlone at MIT. I will be working on problems related to perception and autonomous systems.

Another interest of mine that I am pursuing is thinking about reforms in our education system. Going through it, I can say that I have suffered enough, and I would like to help implement changes. I am looking for means by which I could be of help to Goans who are currently in school and colleges, especially those who are looking for a career in research.

I also write Konkani poetry and my first book on Konkani poetry is set to be released the next time I visit Goa. A plan to release it this June was postponed due to the coronavirus situation.