Robotics is the intelligent association of perception to action. This categorization opens the Proceedings of the First International Symposium on Robotics Research, edited by Brady and Paul (1984). It is remarkable in that it distinguishes the field of robotics without enforcing to define the term robot itself; it covers all the typical work with both mobile and assembly robots as well as a number of unusual limiting cases.
It also suggests, as was the almost general view at the time but is now contentious, that intelligence, perception, and action are somehow distinguishable into distinct functional components of the system. Home Robotics is a young field of study in which dreams proliferate; perhaps, though, the most common of these, both inside the field and out, are that of the sovereign general-purpose robot, able to perform capably in a variety of situations without the need for reprogramming or wide-ranging supervision.
This dream is of both commercial and scientific import: thriving, versatile, inexpensive robotic systems would have an incredible economic and social impact, making it possible to ease people of dangerous, physically demanding, or droning jobs (and, of course, forming the problem of what those people would then do); evenly, the construction of such a robot would give and need substantial insight into how the many diverse kinds of natural creatures are exclusively suited to their own ecological niches, by teaching us about the relationships between environment, agents, behavior, and tasks.
One of the remarkable features of modern robotics, mainly in relation to autonomy, is a propensity to view the problem of constructing a capable autonomous robotic system as one of performance. A robot is self-sufficient when it is able to, and does, takes responsibility for the consequences of its own action while there is no interventionist rescuer offered to fix the problems it manages to form for itself. The robot is thus faced with a diversity of possible situations it might encounter and a collection of possible responses to those situations.
The dilemma of autonomy has to do, then, with choosing proper sensory and action patterns efficiently a question of the performance of the robot in reaction to the challenges offered by its environment. If this were the whole story, home robotic systems might in fact be raised in the near future. However, the rationality of the performance viewpoint covers up the essential difficulty with self-sufficient robotics: The robots are artifacts, and artifacts should be designed. The central questions of designing a system competent of interacting with its environment without recourse to rescue remain to be answered.
Lacking the basic design rules–and design rules entail a well-understood technology and methodology–it is unattainable to answer in prospect rather than in retrospect the key question of whether a particular robot design will congregate the requirements of a given task in a given environment for any state of interesting complications. Certainly, this does not mean that we cannot build robotic systems, only that we run the risk of being surprised by their performance when they attain production.
The current state of knowledge, in comparison with autonomous robotic systems, is rather like that obtained throughout the early years of cathedral construction: The experiential design rules for the towering Gothic buildings were contingents in the course of several centuries of success and failure, and their design was an instinctive insight on the part of the architects concerned until a theoretical analysis of stone structures ultimately became possible.
However, numbers of beautiful (and enduring) buildings were effectively built. One of the key realistic problems of intelligence, then, is the orchestration, on diverse levels, of thought and action. Our ordinary, daily experience illustrates this well. Several of the jobs we do need sustained conscious concentration for success; some happen almost completely subconsciously. There is also a steady migration between these two ends of the spectrum toward the subliminal end. For instance, consider driving.
The majority good drivers spend time thinking about what is going to happen a short time ahead to a certain extent than what is going on now; they do not require to spend attention on changing gear, influencing the pedals, demonstrating their intentions or even, perhaps, on watching what is happening; they mechanically attend to the pertinent details on the road to support their thinking. Much of their skill is subliminal (but not, therefore, unintelligent). The pace and glibness of their reaction to conditions on the road are an outcome of this automation.
On the contrary a novice driver spends attention directly on controlling the vehicle managing the pedals, changing gear, and watching in the mirror all need conscious supervision at the outset. This results in a slower, more arduous performance than that of the expert. The same can be said of numerous activities, mainly (but not only) those involving a considerable element of physical skill: dancing, playing a musical instrument, most sports, and so on. The expert no longer has to listen to the detail of physical performance, and can deliberate on the tactics and strategy of the activity on a good day.
On a bad day, everything becomes bogged down in the details again. The conclusion we might draw from this discussion is that, in human skill, the orchestrations of activity, in which parts of the system are accountable for each component of the skill, differs at diverse degrees of expertise, and the various probable patterns of responsibility for actions have resultant advantages and disadvantages with a revere to the competence as a whole. Precisely the same situation arises in home robotics, excluding that we see the problem from the designer’s viewpoint somewhat than the observer’s.
The design problem is to prefer the orchestration of activity within the robot in such a way that the desired competence is attained robustly, efficiently, and cheaply by the complete system. A good orchestration depends both on the consumption of activity and on the agents (the “players”) themselves. What makes the predicament interesting is that we have a number of commonly interacting levels on which to address the design and no common understanding of the tradeoff involved in preferring any particular level or type of agent with which to progress. A well-understood instance will make clear this point.
Consider the peg insertion task in which a tightly fitting peg is to be inserted into a hole by a robot. The key complexity to overcome is jamming: If the robot tries to force the peg into the hole out of placement, the forces acting on the peg and hole around the points of contact will lean to jam the peg tight. Design Problem There are four qualitatively diverse levels at which we can address this problem: Pegs and holes could be fitted with chamfers (and suitable compliance in the insertion process), so that they auto-locate as the peg is pushed home. In this case, the dilemma is eliminated by engineering the environment.
Chamfers on holes and pegs, part feeders, and careful work-cell engineering are all examples of this approach. The robot could be fitted with a remote center compliance device (Nevins, J. L. , & Whitney, D. E. 1978) that changes the pattern of forces on the peg to prevent jamming. This lessens the problem by an alteration of the robot’s morphology. The robot controller could be amplified to permit compliant motions of the end effector. These third option consequences in the addition of a new class of controlled motions of which the robot is capable, the room of possible actions has been altered.
The robot could follow a cautiously crafted “intelligent” strategy in probing for the hole and inserting the peg. This solution corresponds most closely to the human style of problem solving by means of conscious, strategic, consciously directed activity. This range of possibilities for addressing a difficulty illustrates a general point: The terms in which a problem is depicted need not be directly interrelated to the manner in which its solution is implemented. Particularly, we might describe a problem overtly in symbolic or mathematical form, while solving it using a simple mechanical machination.
The error of implementing the description, though resulting in correct performance, is perhaps the single biggest cause of wasted effort (both human and machine) in robotics. In research into the engineering of human competence, this is also significant: It is unsafe to suppose that the reasonable story advanced by an individual asked how (or even why) they do somewhat bears a close relationship to the truth of how the task is in fact accomplished; we are hardly ever at a loss for an intellectually fulfilling rationalization of our actions.
For robotics, however, a rationalization is insufficient. We require being able to identify the different approaches to a given predicament and enumerate the tradeoff between them in terms of cost, performance, ease of engineering, generality, and so on. In the peg-insertion problem, a few of the tradeoffs are obvious. It might not be reasonably viable or it may be unrealistic for some reason to engineer the environment to suit the robot. Adding extra hardware might limit irrationally the load-bearing capabilities or the physical dexterity of the robot.
Making the controller more multifaceted, to add qualitatively new motor capability to the robot, might be either impossible (the robot uses a proprietary controller that cannot easily be modified) or consequence in an unacceptable performance penalty (the controller now cycles too slowly). The planned strategic peg-insertion might be frustratingly or uneconomically slow, or may rely on repeatability, which cannot be engineered into the robot. On the other hand, any or all of the four approaches, or a suitable permutation of them, might solve the problem in a repeatable, efficient, and economic way.
Cautiously conducted research into skilled human performance can give considerable insight on this kind of question. For instance, consider the competent driver’s capability to brake in such a way as to stop close behind another vehicle or at a traffic light. A psycho physical study by Lee (1986) shows that human performance is well accounted for by the theory that brake timing and forces are controlled using the time-to-contact parameter–a visually-derived estimate of how long it will be until the observer strikes the object being looked at, computed on the assumption of constant relative velocity.
This is astounding as one might assume that successful braking requires knowledge of the distance to the target and of the vehicle’s speed and acceleration, together with some simple reasoning; in fact, the time to-contact is independent of the size, virtual speed, and relative distance of an object (the respective dependencies cancel out) and can be computed solely on the base of image appearance and its rate of change. For a longer discussion of a greater diversity of tasks thought to be dependent on time to-contact and similar visually resultant parameters. Although Lee’s result is primarily surprising, on reflection it is perhaps obvious.
It makes good engineering sense to take to bear a source of information that is easy to access, consistent, and robust, and of general applicability. A single general mechanism might perhaps support the variety of human and animal skills involving the visual cuing of timed action. The psycho physical experimentation, by illuminating an interesting choice in the orchestration of these skills, both illuminates the design tradeoff to some extent, makes doable further theoretical analysis of the proposed model, and suggests useful experimental applications of the theory in robotics.