Environmental Engineering Reference
In-Depth Information
of software agents, the complex constraints placed on them by their physical
environment have forced robot designers to augment the technologies used in
software agents or to develop completely different architectures.
The sensors used by robots measure physical quantities such as environ-
mental image properties, direction and speed of motion, and effector tactile
feedback. They have many attributes that mean complexity in robot designs.
The complex nature of the physical world makes exact sensor measurements
di
cult. Two readings taken moments apart or readings taken by two “iden-
tical” and “healthy” devices often differ. Physical sensors can fail in many
ways. Sometime the failure will result in a device that will give correct results
intermittently. Some sensors require complex processing and in many situa-
tions, the information is dicult to interpret. A vision system using even an
off-the-shelf charge-coupled device (CCD) array can easily supply millions of
bytes of image data every second. Image processing techniques must be used
to examine the data and determine the features of the image relevant to the
robot control system.
Actuators are used to make changes in the physical world. Examples are
opening a valve, moving a wheel, or firing an engine. Like sensors, actuators
can fail in complex ways due possibly to design deficiencies, wear and tear, or
damage caused by the environment. The designers of actuators must also deal
with complex interactions with the rest of the robot and the environment. For
example, if a robot arm and its cargo are heavy, then actions that move the
arm will also apply torque to the robot body that can significantly affect the
sensor readings in other parts of the robot and can even alter the position of
the robot on the supporting surface.
Since both sensors and actuators have complex failure modes, robust sys-
tems should keep long-term information on the status of internal systems and
develop alternative plans accounting for known failures. The unexpected will
happen, so robust systems should actively detect failures, attempt to deter-
mine their nature, and plan alternative strategies to achieve mission objec-
tives. Some systems reason on potential failures during planning in an attempt
to minimize the effects that possible sensor or actuator failures could have on
the ultimate outcome.
Robots exist in a world of constant motion. This requires the robot to
continually sense its environment and be prepared to change its plan based on
unexpected events or circumstances. For example, a robotic arm attempting
to pick up an object in a river bed must be prepared to adapt to changes
in the object's position caused by dynamically changing river currents. Many
robotic systems use reactive control systems to perform such low-level tasks.
They continually sense and analyze the environment and their effect on it
and, within constraints, dynamically change their strategy until the objective
is achieved. The realities of reactive control systems often make them interact
poorly with the slower and symbolic, high-level control systems.
Because of their mobile nature, many robots commit a large percentage of
their resources to navigation. Sensors must support detection, measurement,
