A Duke University graduate has developed a new technology that helps robots to make sense of 3D objects in a richer and more human-like manner.
Autonomous robots have come a long way: Today they are used for inspecting nuclear power plants, cleaning up oil spills in the ocean, accompanying fighter planes into combat and exploring the surface of Mars.
Yet they lag behind when it comes to executing simple tasks such as turning the stove on, fetching the kettle and finding the milk and sugar. While it’s relatively straightforward for robots to “see” objects with cameras and other sensors, interpreting what they see, from a single glimpse, is more difficult.
Now Duke University graduate student Ben Burchfiel and his thesis advisor George Konidaris, an assistant professor from Brown University, have developed a new robot perception algorithm that can simultaneously guess what a new object is, and how it’s oriented, without examining it from multiple angles first. It can also “imagine” any parts that are out of view.
According to Burchfiel, the new approach makes fewer mistakes and is three times faster than the best current methods.
The algorithm was initially tested on a dataset of roughly 4,000 complete 3D scans of common household objects: an assortment of bathtubs, beds, chairs, desks, dressers, monitors, nightstands, sofas, tables and toilets. Each 3D scan was converted into tens of thousands of little cubes, or voxels, stacked on top of each other like LEGO blocks to make them easier to process.
When a robot spots something new – say, a bunk bed – it doesn’t have to sift through its entire mental catalog for a match. It learns, from prior examples, what characteristics beds tend to have. Based on that prior knowledge, it has the power to generalize like a person would – to understand that two objects may be different, yet share properties that make them both a particular type of furniture.
The algorithm was also capable of recognizing objects that were rotated in various ways, which the best competing approaches can’t do. The team is now working on scaling up their approach to enable robots to distinguish between thousands of types of objects at a time.
Image credits and content – Boston University: Neuromorphics Lab/Duke University