Robots and AI can be incredibly interesting to some but daunting to others. Thanks to remarkable progress in the fields of technology and machine learning over the past few decades, AI is now accessible to the masses in ways that weren’t considered possible outside of a science fiction movie.
Manas Joshi, a rising sophomore majoring in computer sciences and data science, has gotten the chance to research it first-hand. He’s one of 32 students taking part in the fifth year of the Letters & Science Summer of Excellence in Research (LASER) program, which is designed to expose students to research under the guidance of experienced professors, faculty and graduate students.
“I’d really like to be a part of something that can help make an impact,” Joshi says. “Whether it’s working in research or whether it’s working in an industry, through the work I’m doing, I’m able to see the impact that I’m making.”
Joshi’s summer research is based in the People and Robots Laboratory in the Department of Computer Sciences. His research involves trying to make robots easier for humans to interact with, a main goal of the lab.
His research is supported by three people in the Department of Computer Sciences: Bilge Mutlu, a professor and the director of the People and Robots Laboratory; Arissa Sato, a postdoctoral researcher; and Callie Kim, a fourth-year PhD student.
“We want to integrate robots naturally into human lives,” says Kim.
Their research focuses on how robots attempt kitchen-based tasks, a process that involves studying how robots understand human intention, process commands and plan the steps in carrying out the instructions. They also monitor how robots understand the concept of safety and take precautions to avoid mishaps.
They use AI in their research to help communicate with robots more effectively. AI is used mainly to interpret human commands into lower-level instructions a robot can understand. They also use AI for object detection in the robot, so the robot can locate and interact with an object as instructed.
“For this user study, we’re trying to understand how people give instructions to others,” says Kim. “How can we model that so that the robot can understand what the human intention was?”
One example of a task a robot can complete is one with which we’re all familiar: making a peanut butter sandwich.
The robot hears this command and interprets it, planning the steps to complete the task. The robot would think to first pick up the bread, stick a knife into some peanut butter, twist the knife and spread the peanut butter onto the bread.
The researchers use data collected from human hands completing the same task for the robot to imitate, making the robot’s movements more lifelike.
The main goal of Joshi’s research is for humans and robots to interact with each other more effectively.
“There’s definitely a gap between how a robot is moving and the human’s expectation of how the robot should behave,” Kim says, “That’s why we want to develop those kinds of interactions, so robots could be more integrated into our daily lives and support people.”
Joshi and Kim are also excited by the prospect of robotics being used to assist humans in real-life contexts, like helping people with disabilities complete a usually difficult task more easily.
The LASER Symposium, where Joshi presented his research on Aug. 8 in the North Atrium of the Chemistry Building, wasn't the first time his name was published in a research format. Going into LASER, Joshi already understood the fundamentals of how to break questions into smaller bits and go through each part to find the answers.
Joshi participated in the College of Letters & Science’s Undergraduate Research Scholars (URS) program for an entire school year prior to beginning his work with the People and Robots Laboratory through LASER. His research with URS was about atmospheric sciences and machine learning and was supervised by Mayra Oyola-Merced, an assistant professor and the Ned P. Smith Distinguished Chair of Meteorology, Atmospheric and Oceanic Sciences.
What inspired Joshi to enter the world of computer science in the first place was when he decided to make a video game.
He first began to learn coding languages and became interested in how they worked. His interests in problem-solving and logical thinking, which he believes are two of the most important aspects of computer science research, began to take root.
“Whether it’s research, whether it’s industry, both always aim to solve problems,” Joshi says.
“There’s a lot of logical thinking that you have to apply to computer science, which is something I’m really interested in.”
This is why Joshi’s research is so important: He is one of many talented researchers who use their vast logical thinking and problem-solving capabilities to understand prominent, futuristic topics like AI and robotics, working out problems and making them less daunting to the masses. One thing that surprised the other members of Joshi’s research team is people’s reactions to their research about robots.
“[People] are kind of scared of robots taking over the world,” Joshi says.
A lot of the comments Joshi and Kim field about their research tend to be positive, yet some still fear a robot apocalypse, citing that robots are becoming too powerful — especially with the rise of AI and machine learning.
So, will robots actually take over the world?
“We are sure that’s not going to happen,” Joshi says. “You can just unplug the cord, which stops the code, and it’s not going to work again.”
“The robot would have to be perfect, and you would have to overcome some obstacles to get there,” adds Kim.
Joshi and Kim agree that people have high expectations about robotics because of sci-fi movies. When these people interact with a robot, they realize that it’s good at doing basic tasks and that it doesn’t have the capability of taking over much of anything.
AI can answer if it has the right data, but it can't sense the world on its own. If someone wants to know the weather, AI wouldn’t be able to answer unless it had something to analyze and use to generate a response; it’s unable to detect the weather itself. And even with data, AI can still make mistakes or "hallucinate," giving confident but misleading answers.
“There’s a lot of persistence you have to have here, because a lot of issues can come up. There are a lot of learning curves you have to overcome,” Joshi says. “There are a lot of things I want to learn, a lot of things I want to understand and a lot of questions I want to ask.”