The algorithm, developed by Massachusetts Institute of Technology researchers and discussed in a paper released Monday, allows robots to gather multiple perspectives of an object, quickly aggregate those images and then use that information to identify the object, according to the school.
But don't expect the algorithm to help a robot clear plates and glasses from a table just yet, said Lawson Wong, a graduate student in electrical engineering and computer science, and the paper's lead author. "As it is now, it's still very far from commercialization," he said.
Improving object detection is just one step in equipping robots to complete house work.
For robots to perform useful tasks in the home, they have to know more than simply how many cups and plates are on the table, he said. If a robot was being used to prepare a meal, for example, it would also have to know what temperature to cook the food or where to find the recipe's ingredients.
Still, the algorithm could eventually help software better compute changes that occur in a home when people move objects and add or remove items.
"The software we use doesn't allow us to capture objects that move over time," said Wong.
Multiple-perspective algorithms allow a robot to identify up to four times as many objects than is possible using a single perspective, and these algorithms also help reduce mis-identifications, according to the researchers.
"If you run [images] through a standard view detection, you will miss a lot of objects," said Wong.
The algorithm also successfully addressed a downside of the multiple-perspective approach: that it can prove time-consuming because it increases exponentially the number of calculations the robot must make, often preventing the robot from completing tasks quickly enough.
The researchers noted that object detectors frequently fail although object recognition is one of the most researched topics in artificial intelligence.