Movie robots are known for helping (or hindering) humans, but in the real world, they have pretty limited social skills. MIT's CSAIL computer science researchers are trying to fix that by teaching robots how to interact with other robots to to further their own goals, according to a new paper. The research could lead to improved human-robot interactions in assisted living facilities, for instance, and even help psychologists better measure social interactions between humans.
To study these interactions, the researchers created a simulated 2D environment that allowed virtual robots to pursue both social and physical goals. For instance, a physical goal might be to navigate to a tree at a certain point on a grid, while a social goal is guessing what another robot is trying to do and then acting based on that, "like helping another robot water the tree," according to CSAIL.
The robot is rewarded for actions that get it closer to its goals, with a matching reward for helping and an opposite reward for hindering it. The team created three types of robots: The first has only physical goals, the second has physical and social goals, but assumes all robots only have physical goals. The third one assumes the others all have social and physical goals, so it can take more advanced actions like joining with others to achieve a goal.
Even young infants seem to understand social interactions like helping and hindering, but we don’t yet have machines that can perform this reasoning at anything like human-level flexibility.
The team created 98 different scenarios with all three types of robots. Twelve humans watched nearly 200 video clips of the robots interacting, and then had to estimate the physical and social goals. "In most instance, their model agreed with what the humans thought about the social interactions that were occurring in each frame," the researchers said.
The researchers hope that the results will act as a "benchmark" that allows others to work on similar social interactions. Next, they plan to create more complex environment with 3D agents that allows more types of interactions. The eventual aim is to not just teach robots how to interact better socially, but "dig deeper into the human aspect of this," said senior author Andrei Barbu. "Can we make an objective test for your ability to recognize social interactions? Maybe there is a way to teach people to recognize these social interactions and improve their abilities."