Page 301 - Introduction to Autonomous Mobile Robots

Chapter 6
286
Local path planning is performed by NF1. The resulting path is converted to an elastic
band that does not take into account kinematics, taking advantage of the fact that the octag-
onal robot used in the exhibition can turn on the spot most of the time. This keeps path
updates as simple as possible. An enhanced dynamic window then takes care of moving the
robot along the path.

6.2.2.8 Nearness diagram
Attempting to close a model fidelity gap in obstacle avoidance methods, the nearness dia-
gram (ND) [107] can be considered to have some similarity to a VFH but solves several of
its shortcomings, especially in very cluttered spaces. It was also used in [108] to take into
account more precise geometric, kinematic, and dynamic constraints. This was achieved by
breaking the problem down into generating the most promising direction of travel with the
sole constraint a circular robot, then adapting this to the kinematic and dynamic constraints
of the robot, followed by a correction for robot shape if is noncircular (only rectangular
shapes were supported in the original publication). Global reasoning was added to the
approach and termed the global nearness diagram (GND) in [110], somewhat similar to the
GDWA extension to the DWA, but based on a workspace representation (instead of con-
figuration space) and updating free space in addition to obstacle information.

6.2.2.9 Gradient method
Realizing that current computer technology allows fast recalculation of wavefront propa-
gation techniques, the gradient method [89] formulates a grid-based global path planning
that takes into account closeness to obstacles and allows generating continuous interpola-
tions of the gradient direction at any given point in the grid. The NF1 is a special case of
the proposed algorithm, which calculates a navigation function at each timestep and uses
the resulting gradient information to drive the robot toward the goal on a smooth path and
not grazing obstacles unless necessary.

6.2.2.10 Adding dynamic constraints
Attempting to address the lack of dynamic models in most of the obstacle avoidance
approaches discussed above, a new kind of space representation was proposed by Minguez,
Montano, and Khatib in [109]. The ego-dynamic space is equally applicable to workspace
and configuration space methods. It transforms obstacles into distances that depend on the
braking constraints and sampling time of the underlying obstacle avoidance method. In
combination with the proposed spatial window (PF) to represent acceleration capabilities,
the approach was tested in conjunction with the ND and PF methods and gives satisfactory
results for circular holonomic robots, with plans to extend it to nonholonomic, noncircular
architectures.