Empower robots with social skills

Robots can deliver food on college campuses, and they can hole-in-one on golf courses, but even the most advanced robots cannot perform the basic social interactions that are vital to human daily life

Researchers at the Massachusetts Institute of Technology have now incorporated certain social interactions into the framework of robots, enabling machines to understand what it means to help or hinder each other and learn to perform these social behaviors on their own

The researchers also showed that their model creates realistic and predictable social interactions

Allowing robots to show social skills may lead to smoother and more active human-computer interaction

"Robots will soon live in our world.

Co-lead author Ravi Tejwani, who is co-lead author Ravi Tejwani, is a research assistant at CSAIL; co-lead author Guo Yanling, a PhD student at CSAIL; Shu Tianmin, postdoctoral fellow in the Department of Brain and Cognitive Sciences; senior author Andrei Barbu is a research scientist at CSAIL and CBMM

A social simulation

To study social interaction, the researchers created a simulated environment in which robots move in a two-dimensional grid, pursuing physical and social goals

The physical goal is related to the environment

Researchers use their model to specify what the robot’s physical goals are, what its social goals are, and how much it should value in this regard

"We have opened up a new mathematical framework for simulating the social interaction between two subjects

Combining the physical and social goals of robots is very important for creating real interactions, because the extent to which humans can help each other is limited

Researchers used this mathematical framework to define three types of robots

Evaluation model

In order to understand how their model compares to the perspective of human social interaction, they created 98 different scenarios with the robot at level 0, level 1, and level 2

In most cases, their models are consistent with human perceptions of the social interactions that occur in each frame

"We have this long-term interest, not only to build computational models for robots, but also to dig deeper into the human aspects of this aspect

More mature

Researchers are working on developing a system with 3D agents that can carry out more types of interactions in an environment, such as operating household items

The researchers also want to add a neural network-based robot planner to the model, which can learn from experience and execute faster
.
Finally, they hope to conduct an experiment to collect data on characteristics that humans use to determine whether two robots are interacting socially
.

Babu said: "I hope we can have a benchmark so that all researchers can study these social interactions, and inspire the scientific and engineering progress we see in other fields such as object and motion recognition
.
"

This research was supported by the Brain, Thought, and Machine Center, National Science Foundation, MIT CSAIL System Learning Program, MIT-ibm Watson Artificial Intelligence Laboratory, and the US Department of Defense Advanced Research Projects Agency Artificial Social Intelligence Success Team Program, the U.
S.
Air Force Research Laboratory, the U.
S.
Air Force Artificial Intelligence Accelerator, and the Office of Naval Research
.

###

Author: Adam Zewe, MIT Press Office

Thesis: "Social Interactions as Recursive MDPs"

https://openreview.
net/pdf?id=3HZLte8gMYS

Article title

"Social interaction as recursive MDPs"