A car-like robot has the ability of obtaining the best estimation of its present position and direction, or its "frame," in the global coordinate system. This functionality is called odometry in mobile robotics. The robot computes present "frame" from the previous one using the wheels’ incremental rotations during the motion sampling time.
A well-known fact in robotics is that this odometry can cause errors to accumulate over time. Odometry errors can be caused by slippage between wheels and the floor, wearing of tires, inappropriate estimation of contact points of tires at the floor, and so forth. However, such errors do not keep MotionMind/Swan from performing its expected funtion.
An example of an odometry error is seen in the trajectory
of Race Track: Dynamic
page). The east-bound and west-bound path segments should coincide because they are constrained by the same side walls. Swan thus believes
it is at a different position from actual position. For most motions, odometry errors do not matter much because they do not disrupt expected functionality. For the motions in which odometry errors do cause problems, one way of resetting odometry errors for a robot would be to equip it with gyros or other absolute position/direction-measurement devices. Swan, however, is not equipped with such equipments. Rather, it depends on a concept called localization, which resets odometry errors using spatial knowledge.
Knowledge-based motion (or spatial-understanding function) is a motion in which Swan uses geometrical knowledge in planning motion. Watch the video of Vacuuming Execution
(Mapping and Vacuuming
page) and Map-Based Navigation (Mapping and Navigation
. In these motions, Swan’s motion depends on a given map, and presence of odometry errors can cause contradictions in operation. Thus, in these motions, we apply algorithms called localization
that reset odometry errors using spatial knowledge. Swan dynamically selects an appropriate localization algorithm for each situation. The algorithm functions in real time to forcibly update the odometry value to a more realistic value. Without localization algorithms, these motions could not be completed correctly.
Localization algorithms can also be applied in several Motion-Reproduction
motions because MotionMind/Swan already has the necessary spatial information. Watch the Pull-Push
function page. The Back-In Parking
motion also applies localization to obtain high-quality motions.
Localization algorithms are the most complex computation in Symmetric Geometry
. This technology will become increasingly more important when car-like robots are used for real-world task execution.