With just a current-gen Kinect camera and some newly created software, the research team was able to create a robot that builds accurate three-dimensional maps of its surroundings, navigable and corrected for errors.
One of the major sources of error in the past has been a phenomenon called “drift,” which results when small errors in the robot’s estimation of its own location compound over many hundreds of different measurements. The results can be devastating to any attempt to actually move through a space, making doorways appear out of line with the stairs behind them…
The team from MIT decided to try to fix the positioning problem by calculating the robot’s change in position between each of Kinect’s 30 frames per second. It uses the data collected from the range-finding camera to build up a 3D model of the environment, then uses the positional information to automatically bend the model into a consistent whole.
This is useful for a number of reasons. It lets a robot build up a sense of physical space without the need for special markers, allowing, for instance, a next-gen Roomba to return to a rest spot intelligently. Combined with input analysis software, however, this could also give a robot the ability to venture into new areas with confidence; Ava500 can already deal with unexpected human movements within a known environment, and combining the two awareness technologies would result in an unprecedented level of autonomy.