maintaining
complex records and expending great efforts in constructing detailed
plans of action, merely apply simple rules for interacting with the world to guide
their behavior moment by moment. Proponents of reactive robotics argue that
when planning a long trip by car, humans do not make all-encompassing,
detailed plans in advance. Instead, they merely select the major roads, leaving
such details as where to eat, what exits to take, and how to handle detours for
later consideration. Likewise, a reactive robot that needs to navigate a crowded
hallway or to go from one building to another does not develop a highly detailed
plan in advance, but instead applies simple rules to avoid each obstacle as it is
encountered. This is the approach taken by the best-selling robot in history, the
iRobot Roomba vacuum cleaner, which moves about a floor in a reactive mode
without bothering to remember the details of furniture and other obstacles. After
all, the family pet will probably not be in the same place next time.
Of course, no single approach will likely prove the best for all situations. Truly
autonomous robots will most likely use multiple levels of reasoning and planning,
applying high-level techniques to set and achieve major goals and lower-level
reactive systems to achieve minor sub-goals. An example of such multilevel reasoning
is found in the Robocup competition—an international competition of
robot soccer teams—that serves as a forum for research toward developing a team
of robots that can beat world-class human soccer teams by the year 2050. Here
the emphasis is not just to build mobile robots that can “kick” a ball but to design
a team of robots that cooperate with each other to obtain a common goal. These
robots not only have to move and to reason about their actions, but they have to
reason about the actions of their teammates and their opponents.
Another example of research in robotics is the field known as evolutionary
robotics in which theories of evolution are applied to develop schemes for both
low-level reactive rules and high-level reasoning. Here we find the survival-ofthe-
fittest theory being used to develop devices that over multiple generations
acquire their own means of balance or mobility. Much of the research in this
area distinguishes between a robot’s internal control system (largely software)
and the physical structure of its body. For example, the control system for a
swimming tadpole robot was transferred to a similar robot with legs. Then evolutionary
techniques were applied within the control system to obtain a robot that
crawled. In other instances, evolutionary techniques have been applied to a
robot’s physical body to discover positions for sensors that are optimal for performing
a particular task. More challenging research seeks ways to evolve software
control systems simultaneously with physical body structures.
To list all the impressive results from research in robotics would be an overwhelming
task. Our current robots are far from the powerful robots in fictional
movies and novels, but they have achieved impressive successes on specific
tasks. We have robots that can drive in traffic, behave like pet dogs, and guide
weapons to their targets. However, while relishing in these successes, we should
note that the affection we feel for an artificial pet dog and the awesome power of
smart weapons raise social and ethical questions that challenge society. Our
future is what we make it.