A new robot that can predict human actions with surprising accuracy is giving restaurant waitstaff and bellhops a run for their money. The Personal Robotics Lab at Cornell University has churned out a machine that can, among other things, refill your coffee cup and hold the door open for you.

This real-world Rosie the robot isn’t clairvoyant — just clever. Unlike simpler robots that can carry out only preprogrammed tasks, this robot observes a person’s movements and makes predictions about what they might lead to next.

By breaking down complicated activities into smaller components, the robot (which still lacks a name) can accurately predict what action a person is performing. So if it sees you reaching, pouring and lifting, it knows you are probably drinking something.

But the robot also predicts what you’ll do even before you do it. This allows it to perform a more useful function. In one laboratory test, for example, the bot observed a person carrying an object toward a refrigerator and helpfully opened the door.

Such thoughtfulness is the result of patient parenting. The machine was taught using a database of 120 3D videos featuring people performing various household activities.

By translating these activities into robot language (i.e. mathematical models), the server robot can identify drinking, eating, cleaning and putting items away. It also associates certain objects with particular activities. [See also: Robots Share Stage With Actors in New Show]

When faced with a new situation, the robot uses its Microsoft Kinect 3-D camera to compare what it observes in the real world to what it learned from the videos.

Even if the actions it sees are slightly different than the model it has on file, the robot still understands what it’s observing and can predict what will happen next. This makes it more useful than robots that blindly carry out a preprogrammed plan.

If, for example, the bot sees you drinking a cup of coffee, it can wait until the appropriate time to refill your cup. The robot knows that when you reach for the cup, you will most likely move it to your lips and then put it back down. This prediction helps it avoid making lap-scalding mistakes.

But don’t fire your flesh-and-blood Jeeves just yet. The further into the future this robot tries to predict, the less accurate it becomes. In tests, it made correct predictions 82 percent of the time when looking one second into the future but only 57 percent of the time when looking 10 seconds into the future.

However, researchers believe that there is a way for the robot to improve its predictions for future actions.