Teaching Robots Correct Etiquette

While workplace politics and etiquette are not something we’re usually officially taught, they’re nonetheless things we inevitably must master if we’re to thrive at work.  As robots become an increasingly common sight in our workplaces, it’s a skill they too must develop so that they can navigate the often complex workplace social environment smoothly.

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have been working to do just that via a new framework they refer to as CommPlan, which provides robots with some high-level principles to guide how and when they communicate.  The robots are then to complete their task as effectively as possible within these heuristics.

The system uses learning and planning algorithms to perform cost-benefit analyses of each decision in real time.  The robot uses this system to weigh up a combination of factors, including gauging whether the human is busy or whether they will respond positively or not.

Optimal approach

In total, a handcrafted policy and acommunications-free silent policy  were also tested alongside CommPlan, with the results showing that the human-robot teams performed both more efficiently and safely when using the CommPlan approach.  Not only is the approach effective, but the team also believe that it’s more time efficient than the other methods, as handcrafted policies typically require a lot of time and expertise to create.

“CommPlan combines the power of human experts and algorithms to create policies that are better and at the same time require reduced developer effort,” they explain.

What’s more, they believe that many handcrafted policies can over-rely on very black and white rules, which can render if harder to cooperate effectively, and often result in overkill and inefficient communication.

“Many of these handcrafted policies are kind of like having a co-worker who keeps bugging you on Slack, or a micromanaging boss who repeatedly asks you how much progress you’ve made,” the researchers explain. “If you’re a first responder in an emergency situation, excessive communication from a colleague might distract you from your primary task.”

They’re confident that the initial strong performance of CommsPlan should render it applicable in areas as diverse as healthcare and manufacturing.  They’re also confident that while the original research only examined the spoken word, it could also work in areas such as augmented reality and the use of visual communication.

“This work is exciting because it reasons about what the human needs from the robot, and the robot is explicitly trying to communicate just the right amount,” the team conclude. “This will enable robots to be more sensitive and responsive to human needs and hopefully make them more helpful to people.”

Facebooktwitterredditpinterestlinkedinmail