Robots are relatively effective when undertaking identical repetitive movements, like carrying out the same task on an assembly line – Pick up a box. Turn the container over. Put the container down in a different location. However, robotics is hampered by the poor ability of robots to perceive objects when the robot is attempting to move through an environment.
To overcome this, researchers at University of Illinois College of Engineering, have developed a new type of filter designed to provide robots with enhanced spatial perception. Through this robots should be able to manipulate objects and navigate through space, at the same time, in ways that are more accurate than current technology permits.
The researchers have used 6D object pose estimation to develop the filter. Current technology is based on 3D — which produces location information on X, Y and Z axes (which is relative location of the object with respect to the imaging device). By 6D, this form of pose is intended to provide a far more detailed picture. In other words, 6D pose estimation is the task of detecting the 6D pose of an object, which include its location and orientation.
According to one of the researchers, Xinke Deng this is: “”Much like describing an airplane in flight, the robot also needs to know the three dimensions of the object’s orientation — its yaw, pitch, and roll…”We want a robot to keep tracking an object as it moves from one location to another.”
The following video shows the 6D process in operation:
This is achieved using a particle filter, which uses a multiple visual samples to estimate the position and orientation of an object,, and this provides a greater set of data for the robot to process. Through this the processor that controls the robot can weigh up the value of importance of each piece information relative to each other. This introduces an relent of uncertainty about the distribution of the orientation of an object, which is something that humans are good at evaluating as the move but robots are not.
The research is detailed in a pre-print paper titled “PoseRBPF: A Rao-Blackwellized Particle Filter for 6D Object Pose Tracking.”
