Hi, Has anyone made an analysis of, or acquired experience in the following ...which is more error-prone, as a way of recording the path of a robot, a list of vectors or a list of waypoints? More specifically, If my autonomous robot moved from A to B along an arbitrary jig-zag path, by odometry only, recording its path as it goes, and then attempted to return to A using the data recorded on the way out, would the error in reaching A likely be less if the path was recorded as a series of vectors which could be reversed to return, or a set of waypoints recorded at each change of course?
A vector defines a direction. You need a distance along that vector, to define a waypoint. Storing either the waypoint or the distance and direction enables you to calculate the other.
The important thing is to define the vectors and distances, or the waypoints, as accurately as possible.
How are you determining vector and distance, or location?
I should have defined my terms; As I understand it a vector has a direction and a magnitude, in this case it would be a heading and a distance. A 2D waypoint would be a pair of coordinates representing a point on some grid or other, in my case a Cartesion grid with the positive x-axis lying along the magnetic North, and the origin (0,0) located at the robot’s Home position.(recharging station).
As it moved away from its home position it could record its movements as a series of vectors
e.g 82(cm), @45(deg)
or it could record the turning points
The components of the vectors would be gotten from a magnetic compass and wheel encoder
The waypoints would be calculated from the vectors.
As I write this I have a stronger and stronger feeling that the two methods would be equally error prone since one can be derived from the other, but I also know that “feeling” that you are right is no proof that you are.
As I write this I have a stronger and stronger feeling that the two methods would be equally error prone since one can be derived from the other
Exactly my feelings.
but I also know that “feeling” that you are right is no proof that you are.
In this case, I think you are.
Vectors are good if the path is actually a "zig zag". Less good if the path is a curve.
What are you actually trying to achieve ? Dead reckoning estimation of position ? Or just a record of where the robot has been ?
Mich: read the original post.
Glen: Waypoints from GPS modules are gonna have a fairly constant error in the range of feet to meters. (at least in my experience). Vectors composed of compass heading and pulse count will have some errors from compass dip and mag deflection. My best luck has been with vectors of direction move from object avoidance routine and pulse count for rpm. Both vector methods are subject to undetermined error from slippage. If you are plotting a zigzag out, you should get the best accuracy by back tracking along your plotted course, If you want to take a zigzag back. circular course deviation induced by slippage is gonna kill you over time though...