A task based approach to the issue of redundant robots starts from the
premise that there are obstacles that cannot be removed from the working area
and which therefore must be avoided. This statement produces the requirement for
a device with a certain degree of mobility, and stresses the need to ensure that
the aim is twofold: reach the goal and avoid obstacles. But avoiding obstacles
is not the same objective as keeping as far away from an obstacle as possible;
the primary goal is still to reach the target. In fact humans use soft contact
to reach targets that are at the periphery of their reach. This soft distributed
contact has the effect of smoothing the surface of the object and hence there is
an element of only being interested in obstacle detail at the appropriate scale
to achieve the task. This paper describes a new approach to collision
avoidance based on using a global path finding algorithm, in this case using
Laplacian potential fields, in conjunction with a simple local geometrically
based algorithm for avoiding obstacles and maximising the use of manoeuvring
space in a manner which is not limited by digital computation resolution issues.
This extra technique is in some ways analogous to the human soft contact
approach. Three examples are presented to illustrate the robustness of
the algorithm. In order to be able to compare results with other techniques, an
environment measurement scheme is defined which gives an indication of the
difficulty of the trajectory being followed.
The applicability of on-line contextual calibration to a neural network based monocular collision avoidance system on a UAV
