Multipath due to the robot's body is certainly a design issue. Generally, you try to use highly directional antennas facing away from the robot so that the influence is minimized.
The particle filter implementations for localization use the motion of the robot or panning antennas to integrate the otherwise sparse signals into something more meaningful. This is analogous to using an antenna array. Still, with these implementations (under the best conditions), you're talking about an error of ~20-50cm on average with a pretty high variance -- meaning other sensors are generally required, though our near-field antennas allow you to hone in on the position even better using the same tag.
Meanwhile, local methods (eg. servoing from the difference between two antennas) works quite well. Publication(s) forthcoming. ;)
Comments
A couple of questions... How do you get rid of the signal rebound due metallic presence in the robot?
provides an object a unique ID with extremely small uncertainty, whose location is inferred by lowerlevel sensory processing
How precise is your method to locate the object? Are you using an array of antennas to calculate the position of every RFID?
Multipath due to the robot's body is certainly a design issue. Generally, you try to use highly directional antennas facing away from the robot so that the influence is minimized.
The particle filter implementations for localization use the motion of the robot or panning antennas to integrate the otherwise sparse signals into something more meaningful. This is analogous to using an antenna array. Still, with these implementations (under the best conditions), you're talking about an error of ~20-50cm on average with a pretty high variance -- meaning other sensors are generally required, though our near-field antennas allow you to hone in on the position even better using the same tag.
Meanwhile, local methods (eg. servoing from the difference between two antennas) works quite well. Publication(s) forthcoming. ;)