Page 240 - Sensing, Intelligence, Motion : How Robots and Humans Move in an Unstructured World

PLANAR REVOLUTE–REVOLUTE (RR) ARM 215

What kind of proximity sensing is acceptable? In Chapter 8 we will address
this question in much detail. Briefly, of particular interest is vision sensing.
As discussed in Section 1.2, vision is good when the scene in which the robot
operates is much bigger than the robot itself. Mobile robots usually operate in
this setting: An obstacle is detected way before it appears next to the robot. One
way to approximate this situation formally is to say that obstacles are (almost)
always detected when they are outside the robot’s convex hull.
The situation changes if the size of the workspace is comparable to the robot
dimensions, which is a standard case for an arm manipulator. Indeed, the arm’s
base is fixed, and the arm is expected to reach all areas of its work cell. Obstacles
are almost always within the robot’s convex hull and can appear next to any point
of the robot body. In this situation the advantages of vision sensing for motion
planning are diminished. Obstacles can appear from behind or from the sides,
at any link of the arm. Depending on where video cameras are attached, an
approaching obstacle may be occluded by cables that deliver power or materials
to the arm hand, or by one of the arm’s own links, or by some other piece of
machinery in the work cell.
Trying to fight this problem by covering the arm (or the work cell walls) with
many “eyes” will make the system awkward and hardly feasible. Another option,
decreasing the number of cameras by putting them in a few strategic locations
and then using auxiliary arm motion to disambiguate invisible obstacles, is even
more awkward and creates other difficulties. The conclusion is that vision is
useful in a limited role, such as protecting the arm’s gripper when manipulating
objects. To protect the whole arm body at short distances, sensing media that
provides a physical coverage of the body will likely serve the task better than
vision.
Various proximity sensing devices—infrared, ultrasound, capacitance, and oth-
ers—fit our needs well. All of these have some limited distance of operation,
and all have their own pluses and minuses. For example, ultrasound sensing has
a wide sensing range that will likely reach the boundaries of the arm workspace,
but its resolution is not very good. Other properties—for example, accuracy, reli-
ability, and so on—may be important as well in the choice of sensing hardware.
Without going into more specifics (see Chapter 8 for more detail), let us assume
here that our revolute–revolute arm is equipped with some generic proximity
sensing hardware, such that every point of the arm body can sense approaching
obstacles.

Algorithmic Issues. How do we incorporate proximity sensing in the RR-Arm
Algorithm or similar motion planning procedures? In the discussion on sensing
versus motion planning for mobile robots (Section 3.6), we assumed that the
robot has a circular sensing range, with some limited radius of sensing. For
a mobile robot, this is a reasonable and natural assumption; it can be easily
modified for some practical constraints, such as partial sectoral sensing. For
an arm manipulator the situation is more complex. Assuming a similar limited
sensing distance at each point of the robot body, in the workspace the outline of