Chapter Twelve Robotics The Ultimate Intelligent Agents Defining

Chapter Twelve Robotics: The Ultimate Intelligent Agents

Defining Robotic Function · A mechanical entity that can function autonomously, by which is meant: · Without recourse to a human operator. · Able to adapt to a changing environment. · Continue to function when one of its own less important parts break. · Move within and change its world circumstances.

Historical Highlights · · · · · 400 B. C. : A wooden dove that flaps its wings. 1500 s: Robots that play music 1600 s: More sophisticated mechanisms. Late 19 th century: Remote control. Early 20 th century: Electronic devices are introduced. 1940 s: Industrial telemanipulator for radioactive applications. 1966: Shakey, the first AI robot. 1970 s: Increasingly sophisticated robots for semi-autonomous exploration of remote surfaces. 2000: Lamprey brain is connected to sensors to control a robot. 2005: Duke/Cal Tech scientists explore techniques that will enable humans to operate exclusively through brain signals.

Evaluating Robotic Potentials • For fully autonomous performance approaching human capability, robots would need to: understand speech, see, plan, reason, represent a world model, learn. These are truly awesome accomplishments.

Biological Foundations of Robotic Paradigms The ability to quantify human behavior is a foundation for being able to emulate intelligence. · Lorenz/Tinbergen codify the way in which an animal acquires and organizes behavior. · Starting from a sequence of innate behaviors (e. g. , feeding), new behaviors can evolve (e. g. , hunting is composed of searching, stalking, chasing, etc. ). ·

Evaluation of Lorenz/Tinbergen • Their model fails to provide adequate explanation for dynamic aspects of behavior. It reflects a “top-down” philosophy and does not sufficiently account for perception—a behavioral “releaser. ”

Action-Perception Cycle of Animal Behavior • Neisser/Gibson provide a dynamic model of human behavior. Agent Acts Modifies Actions and Behaviors Interaction With the Environment Perception of World Changes Its Perception (new viewpoint)

Evaluation of the Biological Basis of Robots • Psychologists cannot account for a number of phenomena that need to be resolved before transfer to mechanical intelligent agents: concurrent behavior conflicts, missed affordances (some behaviors may not be described simply by sensory-action activities), learning (not fully resolved among cognitive scientists).

Foundations of Robotic Paradigms A paradigm is a philosophy for working with a class of problems. • Each of the prominent robotic paradigms includes a series of primitive functions: sense, plan, act. • Sense: convert elements of an environment into information used by other parts of the system. • Plan: elements corresponding to human reasoning capabilities. • Act: includes the motor and activation elements of robotic environments. •

Evaluation of Paradigm Foundations • The ability to learn is a biological feature of more advanced animals. A growing number of Roboticists believe that a new primitive needs to be added to robotic architectures—a learn process. There are presently no formal organizations in which such a process is fully integrated.

The Hierarchical Robotic Paradigm Includes sensors and possible feature extraction SENSE Creates a model; develops a plan to complete a task; produces commands for the actuators Controls actuators PLAN ACT

Evaluation of the Hierarchical Paradigm PLAN reflects the way people “think” about an action. However, not all action is “preceded” by thinking. Humans may have a repertoire of default schemes for completing a task. • This model presupposes a single global model of the world. Generic global world models do not handle “surprises” very well. •

Reactive Paradigm (also known as Subsumption) A fundamental behavior SENSE Sensor 1 ACTUATOR Behavior 1 Behavior 2 Sensor 2 Behavior 3 Complex, “intelligent” behaviors—a combination of simple behaviors

Evaluation of the Reactive Paradigm Whether such architectures can be ported (reused) to new applications is an open question. They are not easily transferred to domains where reasoning about resource allocation is essential. • Lack redundancy (e. g. , a second of backup sensing system). • Assemblages of behaviors depend heavily on the programmer. •

The Hybrid Paradigm • • • Designs characterized by a combination of reactive behaviors and planning. The PLAN component includes a deliberative process. Behavior includes reflexive as well as innate and learned behaviors (skills). Assemblages of behaviors sequenced over time, rather than primitives. Planning can include: path planning, map making. Hybrids also include performance modeling.

Evaluation of Hybrid Architectures Full evaluation is difficult because Hybrid organizations are still evolving. • There is no currently predominant architecture; each must be considered in light of its application. • Are Hybrid designs really unique or merely variations of Hierarchical architectures? • Can suffer from limitations of computing capacity and an associated paucity of planning intelligence. •

Overall Evaluation of Robots (Two views from the same institution) • “The body, this mass of biomolecules, is a machine that acts according to a set of specifiable rules. . . I believe myself and my children all to be mere machines” Rodney Brooks, Director of the MIT AI Laboratory • “The reason there are no humanlike robots is not that the very idea of a mechanical mind is misguided. It is that the engineering problems that we humans solve as we see and walk and plan and make it through the day are far more challenging than landing on the moon or sequencing the human genome. Nature, once again, has found ingenious solutions that human engineers cannot yet duplicate. ” Steven Pinker, Director of the Center for Cognitive Neuroscience at MIT