Figure 1: Mobile robot with ultrasonic sensors
Source: Linder, M. Harris-Warren, K. "A Robot Firefighter",
Circuit Cellar Ink. No 63. Oct. 1995. Pp. 14 - 22.
In an ideal environment there exists only one reflector, perfect surfaces, and no variations in temperature. Thus a ultrasonic sensor system can easily send out an ultrasonic wave and that wave will bounce back of an object and return to the receiver an
d then distance can easily be calculated. Yet in a practical world this is not the case. Usually one sensor is not enough; most mobile robots use multiple ultrasonic sensors to overcome the limitations of individual sensors. Typically there are many r
eflectors and multiple echoes. One must wonder then; which echo belongs to which receiver? Furthermore, if one is using all this distance information to create one map, then how can one tie all this information together? The general problem of all of
these situations is known as the correspondence problem.
In the process of mapping an object, there are sometimes areas of ambiguity , in which the sensors may give back fluctuating data. In most cases, these areas cause a correspondence problem.
In another situation, false reflections can cause extremely large errors in a sensor reading. (Refer to the figure 3 below.)
As we can see, figure 3.a illustrates a indoor environment. Figure 3.b illustrates a sensor reading. When we put the two illustrations together (figure 3.c) we can see that due to false reflections, certain walls were perceived to be farther away then they are. This problem can be corrected by building the environment map with multiple sensor readings that are taken from different locations. In similar situations, corners or gaps in the environment may be missing from the computer mapping. In thi s situation, a similar solution can be used.
An autonomous mobile robot must be capable of perception, decision making, planning, navigation and control in order to carry out intelligent tasks. A successful robot will be able to avoid obstacles while navigating through different terrains. During
the past few years, potential field methods (PFMs) and grids methods have been commonly used. Yet there are some significant problems that are inherent in PFMs. A type of grid method developed by Borenstein, known as a vector field histogram (VFH),
uses grids for the representation of obstacles. This method uses a two-dimensional Cartesian histogram grid as a world map. The world model is updated continuously with range data sampled by the on-board range sensors.
The way the grid works is that each cell in the grid represents a certainty value (CV). A probability is projected onto a cell in the grid by a range reading. For ultrasonic sensors, this cell corresponds to the measured distance and lies on the acoust
ic axis of the sensor. A probabilistic distribution is actually obtained by continuously and rapidly sampling each sensor while the robot
is moving. Thus, the same cell and its neighboring cells are repeatedly incremented.
(Refer to figure 4 below for an example of a VFH.)
Figure 4: Grid
Source: Zhao-Qing, M. & Zeng-Ren, Y.
My research focuses on the creation of a real-time map for the mobile robot that our group is working on. This map will be grid-based and is similar to a VFH. My main objective before the semester ends ( May 1997) is to interpret the distance informati on from an ultrasonic sensors into a workable map. This map will be continuously updated as more information from the ultrasonic sensors come in. Later, this map can be used in conjunction with a navigation program to give the robot the ability to move around in its environment.