Architectures for Sensor- Based
Navigation
Within the mobility of the vehicle, the sensor-based motion system is the
part in charge of generating movement free of collisions between
successive positions. The design of these systems is determined by diverse
factors involved in this question, like the model construction, the
deliberative planning and the motion generation. The model builder
constructs a representation which is the base for the deliberation and
which provides with memory the reactive behavior (obstacle
avoidance), the planner module generates global plans
and the reactive module computes the local motion. The sensor-based
systems made up as synthesis of modules with these functionalities mainly
differ in the interaction between the planner and the reactor (i.e. how
the reactive navigation uses the information available of the planner),
and in the tools used to implement each module.
We give in this Section a global vision of the sensor-based system, which
is formed by an architecture that integrates three modules with the
following functionalities: model construction, motion planning and
reactive navigation:
-
Model
Builder Module: construction of a model of the environment (to
increase the spatial domain of the planning and used as local memory for
the reactivity). We use a binary occupancy grid that is updated whenever
a new sensory measurement is available. The grid has a limited size and
travels centred with the robot. Furthermore, we employ a scan matching
technique to improve the vehicle odometry before integrating any new
measure in the grid. Although, a scan matching technique does not
guarantee global consistency, its precision is enough to build the local
map needed by the other modules.
-
Planner
Module: extraction of the connectivity of the free space (used to
avoid the cyclical motions and trap situations). We have developed a new
planner that computes the existence of a path that joins the robot and
goal locations. The planner constructs iteratively a graph whose nodes
are locations in the space and the arcs are tunnels of free space that
joins them. When the goal is reached, the current tunnel contains a path
to the goal. This planner avoids the local minima and is very efficient
so that it can be executed in real time.
-
Obstacle
Avoidance: computation of
the collision-free motion. We chose the Nearness Diagram Navigation (ND
method in short), which is based on selecting at every moment a
navigational situation and to apply a motion law adapted for each one.
This method has demonstrated to be very efficient and robust in
environments with little space to manoeuvre.
Globally the system works as follows (see
Figure): given a laser scan and the odometry of the vehicle, the model
builder incorporates this information into the existing model. Next, the
information of obstacles and free space in the grid is used by the planner
module to compute the course to follow to reach the goal. Finally, the
reactive module uses the information of the obstacles contained in the
grid and information of this tactical planner to generate the motion (to
drive the vehicle free of collisions towards the goal). The motion is
executed by the vehicle controller and the process restarts with a new
sensorial measurement. It is important to stress that the three modules
work synchronously within the perception - action cycle. Next, we address
the design of the modules and the integration architecture.