Decoding Artificial Intelligence with Prof. Gaurav Sukhatme
Artificial Intelligence (AI) is one of the most discussed areas today and it will heavily influence the transformation of businesses. This is an edited excerpt from an interview of Prof. Gaurav Sukhatme, Fletcher Jones Endowed Chair in Computer Science, Executive Vice Dean at University of Southern California and Amazon scholar 2020; conducted by Amit Paranjape, Chairman of the IT & ITES Committee at MCCIA as a part of MCCIA’s YouTube series, ‘Decoding Artificial Intelligence with Industry Leaders’. They talk about the history of robotics, its evolution, ways to make robots more intelligent, and the future of robotics. Prof. Sukhatme also explains the applications of robots in MSMEs and advises students aspiring to build a career in this field.
Prof. Gaurav Sukhatme has done his engineering in Computer Science from IIT-Bombay. His grandfather Pandurang Sukhatme was an award-winning statistician. His father Prof. Suhas Sukhatme was a professor of mechanical engineering at IIT-B and went on to become the director of IIT-B. Owing to his father’s job Prof. Sukhatme grew up on the IIT-B campus. Though he came from an academically-oriented family, he was given complete freedom to choose subjects according to his liking.
After graduation, he went to the University of Southern California to complete his masters and Ph.D. and has been working there for almost 3 decades now. In his early days of research, he focused on developing techniques for multiple mobile robotic systems to coordinate their behaviour with each other. Today, these robot teams are called swarm robotics systems and can be used for structured work on the shop floor for logistics, material handling as well as unstructured work like rescue and rehabilitation in case of a natural calamity.
In 2000, he collaborated with limnologists and oceanographers and developed techniques to put teams of robots into the ocean and lakes. As he worked with people who studied the natural environment he became interested in answering questions that came from outside his lab and had to think of creative ways in which they could invent new techniques and algorithms to answer that. Subsequently, he branched into other domains in robotics. Today his lab works primarily in the intersection of doing learning and planning for various kinds of robots. Learning and planning are in many ways the cornerstones of what is Artificial Intelligence (AI).
Prof Sukhatme says that AI is not new; even in 1989, the students at IIT-B had to take a class on it. But in the early 1990s, the study of robotics was mostly theoretical, while today most leading labs in the world have access to a fair number of robots, and students can apply their ideas and try different things. Certain kinds of simulation technologies and open source have taken off over the last two decades which encourages students to express their creativity, and creativity is the key in research. The exciting work in the field of AI over the last decade is powered by the Deep Learning approach. While computer scientists have tried to get machines to recognize images for 30 years, the algorithms have become smarter over time and today machines have become incredible at image recognition. The same is the case with natural language processing which makes it possible to interact with machines through voice commands.
He explains that robotics is AI for the physical world, unlike disembodied AI where data is pumped in and an answer is pumped out. It is a challenge because interaction has to happen in a closed loop with the world. It requires careful consideration of all the concepts that mechanical engineers and physicists have studied for centuries and hence to see progress on robots that are physically more intelligent is going to take some more time.
Currently, his research is focused on the area of autonomous, self-learning robots. He is trying to develop techniques where robots can be trained to do tasks that are built on top of tasks that they’ve already learned. In simple terms, if a robot can pick a certain object from a floor and put it into a box, it should be able to do this for all objects, irrespective of its shape, light conditions, and other distractions. But, according to him, the challenge is getting the robot to do this by programming it only for some instances and getting it to adapt automatically to the other situations. To solve this problem in the lab, they are exploring 2 techniques- one is the human lead approach which is very popular and involves teaching by showing, and the other is a kind of kinesthetic way, where the human picks up the robotic arm and moves it from place to place. He says that this can be done by buying a teaching pendant with the robot that allows the robot to be driven manually from place to place and then having the robot replicate the trajectory. However, in his, opinion the real excitement is when a new behaviour is generated instead of just being able to replay.
He also elaborates on how AI can be used in the manufacturing sector by the MSME industries. Various low-cost robotic solutions can be used on the shop floor for tasks like packaging and material handling. One of the biggest advantages of AI in this sector is that robots can be retaught with very little reprogramming and hence turn out to be useful over a longer time span even as their tasks change. Another benefit is that robots can be taught a sequence of tasks. So instead of having one machine do the same task repetitively, one machine would be able to perform multiple different tasks making it possible to accomplish complex jobs. These technologies are maturing at a remarkable rate and tremendous advances are likely to happen in the adoption of robots for heavy industries.
When asked about the fear of robots taking over the jobs of people, he says that these fears are unfounded. People have limitless creativity and robots will allow them to use their skills in newer areas and solve higher-level problems. Robots continually improve their performance by proactively asking for feedback when they recognize that they’ve reached the limit of their performance and hence people play an important role in it which makes this technology that is not exclusive of people but one that will work with them
While speaking on the different skill sets required for robotics, he explains that robotics is an integration of different fields. One can contribute to robotics by being a good programmer, by being good at math, by being good at physics, by being good at thinking systemically and thinking about human interaction.
Prof. Sukahtme concludes by saying that in the next 15–20 years AI will be applied extensively in the area of self-driving cars. Though cars will not suddenly become completely self-driving, a lot of technologies will be infused together to provide a higher level of driving assistance. Homes will become more advanced. There will be increased automation available for the home, particularly for better cleaning and security techniques. Robots will be able to talk to people more effectively. Currently, the hardest part is to get robots to be good at handling physical objects, but there will be advancements on that frontier as well. Robotic things will find their way in every facet of life through autonomous technologies that help people.
(You can watch the entire interview at https://www.youtube.com/watch?v=_e2QnIPes9E )