Social Robotics: How Scientists Are Teaching Robots to Teach

Robot learning

A social robot was recently taught to teach by giving encouragement to young students between the age of eight and ten. In order to do so, a group of researchers implemented a new machine learning system: Supervised Progressively Autonomous Robot Competencies (SPARC).

SPARC is a relatively new approach that has previously only been tested in simulations or with other robots standing in for humans. Emmanuel Senft and colleagues developed the methodology and are now improving it with human tests.

The robots used in the study were able to acquire appropriate social behavior for the classroom by interacting with students under close supervision from a psychology Ph.D. student with no background in robotics.

Human control and robot autonomy

In order to adequately adapt the robots' behavior for the classroom, the researchers strived towards striking a balance between human control and robot autonomy.

This challenge is an ethical as well as technical one that is currently unachievable through traditional machine learning methods, like reinforcement learning.

With SPARC, the robot’s control actions are initially a blank slate. The teacher teleoperates the controller and chooses actions for the robot to execute.

In the tests, the robots were shown to catch onto learning patterns quickly. The robots "evolve" and start to suggest actions to the teachers, based on their previous class time and experience. The teacher continues to choose which actions to allow, furthering the robots' learning on what actions to pursue in particular settings. Over time this reduces the need for teacher intervention.

The benefits of social robotics

Using SPARC, an NAO humanoid robot was able to effectively acquire social behavior that allowed it to tutor 75 children. It was able to learn this behavior after only 25 interactions, lasting roughly three hours.

The robots help led to a 10% to 30% improvement in the students’ performance. 

SPARC's potential isn't limited to education. Beyond applications for the classroom, the study's authors say social robotics can help to improve medical and assistive robotics.

Social robots, they say, can help fill a crucial gap in education. A rising number of pupils, as well as budget cuts, have made one-to-one interaction more difficult to come by for students.

Co-existing with robots

Interestingly, as wee see robotics technology become more common in classroom settings, robot abuse, and vandalism has become a problem for some educators.

One group of researchers created a robot that teaches children in a different way — it helps them learn how to behave around robotic technology.

In any case, the SPARC researchers hope their new adaptive method of social behavior learning for robots leads to the technology catching on and becoming a bigger part of our society.

If it does, robotics could be a great aid in helping humans socially. One of the great benefits would be helping young people to become more engaged in education.