What Is Autonomy?
Autonomy refers to systems capable of operating in the real-world environment
without any form of external control for extended periods of time. Thus, living systems are the prototypes of autonomous systems: They can survive in a dynamic environment for extended periods, maintain their internal structures and processes, use the environment to locate and obtain materials for sustenance, and exhibit a variety of behaviors (such as feeding, foraging, and mating). They are also, within limits, capable of adapting to environmental change.
The emphasis on behaviors makes it clear that we do not consider a rock an autonomous system. Clearly, it exists in the world without external control, but it is capable neither of operating in the world nor of exhibiting any behaviors.
The emphasis in this book is on autonomous systems created by humans. Frequently,these systems draw inspiration from biology, but not always. For example,many autonomous systems use wheels for locomotion, and no wheels exist in nature.‘‘Capable of operating’’ implies that these systems perform some function or task.This function may be that intended by their human creator, or it may be an unexpected,emergent behavior. As these systems become more complex, they are likely to exhibit more and more unexpected behaviors.
It should be clear that at the present time, most robots are not fully autonomous,within the scope of the preceding definition. They are not capable of surviving and performing useful tasks in the real world for extended periods, except under highly structured situations. However, if the environment is su‰ciently stable and the disturbances to it are not too severe, robots can indeed survive and perform useful tasks for extended periods. Furthermore, the field of robotics is a very active research area at this time, and we can expect robots to exhibit increasing levels of autonomy and intelligence in the near future. In certain structured situations, for example, the international RoboCup competitions, small teams of robots already exhibit full autonomy while playing ‘‘robot soccer’’ (Asada and Kitano 1999a).
What Is a Robot?
A robot must have sensors, processing ability that emulates some aspects of cognition,and actuators. Sensors are needed to obtain information from the environment. Reactive behaviors (like the stretch reflex in humans) do not require any deep cognitive ability, but on-board intelligence is necessary if the robot is to perform significant tasks autonomously, and actuation is needed to enable the robot to exert forces upon the environment. Generally, these forces will result in motion of the entire robot or one of its elements (such as an arm, a leg, or a wheel).
This definition of a robot is very broad. It includes industrial robot manipulators,such as those used for pick-and-place, painting, or welding operations, provided they incorporate all these three elements. Early industrial manipulators had neither sensing nor reasoning ability; they were preprogrammed to execute specific tasks. Currently most industrial robots are being equipped with computer vision and other sensors and include on-board processors to allow for some autonomy.
Problems of Robot Control
- A robot should never harm a human being.
- A robot should obey a human being, unless this contradicts the first law.
- A robot should not harm another robot, unless this contradicts the first or second
law.
stability. Software architectures allowing for such control processes to proceed in parallel are known as subsumption architectures (Brooks 1986) or behavior-based architectures (Arkin 1998).
The software organization associated with these multiple levels is often termed the control architecture of a robot. We examine these various aspects of control in detail in succeeding chapters of this book, but some of the basic issues are discussed in the following paragraphs. Clearly, the higher levels of control provide inputs to the lower levels, but there is also feedback from the lower levels to the upper levels. Sensors provide inputs to the lowest (and sometimes the intermediate level); actions upon the world are exerted from the lowest level.
Low-level control is clearly autonomous, whereas intermediatelevel control is generally autonomous in contemporary mobile robots but may still involve some human input. As indicated previously, this is an extremely active area of research, and we can expect increasing autonomy even at the highest level. ‘‘Structure shift,’’ referred to in the figure in the context of high-level control, implies an ability on the part of a robot to reconfigure its physical structure; some robots are already capable of some autonomous reconfiguration (see, e.g., Rus and Chirikjian 2001; Shen, Salemi, and Will 2002).
Biologically Inspired Robot Control
Recent research has increasingly emphasized the use of behavior-based strategies for control of autonomous robots (Brooks 1986; Maes and Brooks 1990; Arkin 1998; Beer 1990). One of the major motivating factors behind this approach to autonomy arises from the di‰culties sociated with traditional methods, which require accurate knowledge of the robot’s dynamics and kinematics, as well as carefully constructed maps of the environment in which they operate. Such approaches are not well suited to time-va rying and unpredictable, unstructured situations. As a solution to this problem, Brooks and others have proposed reactive strategies: The robot senses the environment and reacts with appropriate behaviors as required.
Sensors
Robots need sensors both to receive information from the outside world and to monitor their internal environment. Many (but not all) robot sensors are devices that attempt to imitate some of the properties of animal senses. In this section we provide a brief introduction to the major sensory systems in animals, indicate how some of the features of these sensors are incorporated into sensing devices used with mobile robots, and list the major limitations of these devices.
