artigo Beyond Top Down: Designing with Cubelets

Cubelets is a modular robotic building set that embodies a parallel distributed computing model. Unlike central-brain, top-down command-and-control models that conventional robotics construction kits employ, a parallel distributed computing model accounts for emergent phenomena in the world. We compare the top-down and parallel distributed models in the familiar task of constructing a maze-following robot. As we consider how these different robotics models contribute to maze following we also examine what role the parallel distributed model has in influencing students’ success, both at introductory robotics and at developing critical thinking skills.


Introduction: Beyond Top Down
Some decades ago now, Nobel Prize winning economist Thomas Schelling wrote an influential article titled "On the Ecology of Micromotives" (1972).Using a simple cellular automata model to represent the distribution of "white" and "black" residents in a city, he showed how local preferences such as, "I don't want to be the only person of my color in my immediate neighborhood" could produce segregated cities.
Although such a benign preference might initially appear to be inconsequential, in fact Schelling showed that it results in radical segregation.In this model there is no designer causing segregation.It can even be said that no individuals intend to contribute to segregating the city but that is the ultimate outcome.Without explicit design for segregated cities, segregation can still arise from the combined local acts of individuals.
Here is an example of complexity-an unexpected global phenomenon emerges from the interaction of local behaviors.It is difficult to grasp, partly because we are accustomed to thinking about clear and direct relationships between cause and effect.If a city is segregated, surely someone must have designed it so.Not to deny the powerful political and economic forces that do in fact work to maintain segregated cities, Schelling shows that even simple individual decisions made locally can have profound global effects.
We have a bias towards simple, single-action and single-solution thinking.We teach engineers to think top-down: What is the problem?How can we solve it?This is the essence of engineering as we teach and learn it today -it makes a great deal of sense for many problems and so permeates our strategic thinking beyond engineering.As we tackle big problems in our daily lives we follow similar suit and use the "divide and conquer" strategy: What are the parts of the problem?How can we solve each part individually?
And yet, the problems of a new millennium challenge the top down command and control models that carried us to where we are today.On a number of fronts, we are faced with the emergence of global effects of local interactions.Top-down thinking fails to address these situations.We must learn to operate within this new paradigm in order to respond effectively to today's challenges.Advancing our critical thinking and successful problem solving depends on seeing and thinking about the world in profoundly different ways than before.
For example, AI pioneer Marvin Minsky's Society of Mind (1988) argues that thinking is essentially a complex community of agents.In The Blind Watchmaker, Richard Dawkins (1987) shows how the filtered randomness that takes place in evolution accounts for what might otherwise be thought to be a process of intentional design.
In the top down paradigm a designer determines the behavior of a system.The machine works because we designed it that way.This single-cause and command and control perspective influences how we view many kinds of systems -machines, politics, and even ourselves.We think of ourselves as a body controlled by a brain-our eyes and ears tell our brain about the world, and our brain tells our hands and feet what to do.To be sure, our bodies are far more complex, but this simple top-down view dominates our understanding.understand that the world is more connected and complex than we understood before.
We've learned that local perturbations can have massive global effects (the proverbial "butterfly effect").The top-down paradigm will no longer suffice; we must learn to see the world differently (Wolfram, 2002;Johnson, 2001).And we must educate a new generation of scientists and engineers to go beyond top-down thinking.
As we catalog the challenges of the 21st century, we are increasingly concerned with teaching science, technology, engineering, and mathematics (STEM).Our students must become problem-solvers and innovators, and we must prepare them to intervene in a world of complex problems.
To address this challenge, many in-school and out-of-school programs teach robotics.Robotics offer a compelling and scalable instance of interdisciplinary STEM learning, and many of today's young students find robotics competitions engaging and motivating.The Lego FIRST, VEX robotics, and Robot Soccer World Cup competitions are three well-known examples.In these, student teams compete to design and build robots that accomplish a challenging task.Through designing and building robots students learn real-world STEM skills.They learn mechanics, electronics, and programming, and they learn to work together to engineer complex devices that meet a stated goal.
Because of the specific technologies that most robotics education employs, these approaches to teaching engineering assume and reinforce the top down paradigm.Simply put, the robots students build comprise a single central 'brain' computer that controls the robot with a collection of sensors and actuators that mediate between the computer and the physical world.Engineering a successful robot is a matter of constructing the mechanics and electronics and then bringing it to life by writing command-and-control software that runs on the brain computer.Certainly designing and building robots with conventional single-brain central processor kits is an effective way for students to learn important engineering principles.Yet we believe that this approach to teaching STEM through robotics misses an opportunity to teach engineering design outside the top-down paradigm, and to begin to engage young students with designing with complexity in a hands-on way.
Our robot construction kit, Cubelets, breaks the top-down view of the world.It embodies instead, the local-to-global paradigm.Like other robot construction kits, it's in the constructionist tradition of Papert (1980).However, by giving young people experience in how large-scale behaviors emerge from local interactions, we aim to scaffold new ways of thinking about the world.To illustrate the difference, we turn now to a traditional elementary robotics task, to build a robot that finds its way through a maze from start to goal.

The maze-solving task: strategy and specific engineering challenges
A common challenge in beginning robotics education is the maze-solving task.
Students must design and build a robot that autonomously navigates a physical maze.
Figure 1 shows a typical example.task is timed and if left-hand and right-hand walls do not comprise a "trip" through the maze of equal lengths).Specific engineering challenges of building a robot that solves a maze in this way include: • sensingthe distance to the left (or right) wall.
• actingturning precisely while moving to maintain a fixed distance to the wall.
Although solving a maze is an elementary robotics challenge it serves to illustrate the differences between following the usual top-down approach and the distributed computing approach that Cubelets encourages.

Thinking with your hands: Manipulatives Matter
We are certainly not the first to argue for the importance of understanding emergent behavior and complex systems through computational thinking.Perhaps best known is the work of Mitchel Resnick on StarLogo and Uri Wilensky on NetLogo (Resnick, 1994;Wilensky;Resnick, 1993; (see also Bonabeau et al., 1999).In these languages a programmer writes code to control the behavior of a single on-screen 'turtle'.This produces emergent behavior when each turtle in a swarm executes the same code in parallel.For example, a simple follow-your-neighbor program causes turtles to exhibit flocking behavior.
Although these screen-based systems provide powerful platforms for experimenting with emergence of complex behaviors, we believe that a physical instantiation of these phenomena offers significant advantages.
The current fascination with screens-smart phones, tablets, laptop computersreminds us that interactive computer simulations on displays can be convenient and instructive.As compelling as screen based simulations of physical phenomena may be for some, for many others experimenting with physical robots is more engaging.Hands-on learning eliminates a layer of abstraction.When a real-world phenomenon is represented on a screen (e.g., birds flocking), some explanation or schema must accompany it that relates the screen simulation to real-world phenomena.Discovery based, hands-on activity allows students to engage directly and with less explicit instruction.To be sure, some kinds of learning are more safely or thoroughly accomplished through simulations (e.g.flight training), but when students can accomplish learning objectives with physical materials, manipulate objects, and produce results, they have more control and freedom to explore.They also can debug and improve their creations without additional abstraction.Manipulating objects, making and creating, and hands-on learning is in the rich tradition of Dewey (Dewey, 1925), Froebel (see Brosterman, 1997) and Papert (1980) and as we've noted above, others in the constructionist community have addressed the significance of understanding parallel and distributed models of computation.The Cubelets modular robotics construction kit brings these two models together into one tool that is both inviting to children and powerful enough to solve classical robotics tasks.

The top-down model
Typically, robotics students are equipped with a kit or robotics system to address various tasks, such as the maze-solving task we discuss here.For these reasons, dead reckoning is seldom used in building maze-solving robots; indeed, in most competitions, a dead reckoning algorithm would not be considered meeting the maze-solving goal.Most simple maze-solving robots eschew dead-reckoning and opt to either construct either a Linear Time Invariant (LTI) system or a system that includes hysteresis.Both these robot types employ a central processor attached to peripheral sensors and actuators.They differ in that a hysteretic system records past states and data from trials as well as responding to immediate sensory information.
An LTI solution is predicated on the central processor that interacts with peripheral devices (sensors and actuators).The processor operates as a governor that manages rulebased cases in real time.If the peripherals are touch sensors and wheeled actuators, the robot can be programmed to respond to current information such as hitting a wall and correcting, or sensing an opening ("no wall") and turning to enter it.A robot of this type is liberated from merely acting out a stored map, and because it interacts with its environment, it is more effective than the dead-reckoning solution.If its wheels move slower than a mapprogram had anticipated (due to weak batteries, slipping wheels, or other real-world exigencies), the robot is not locked moronically into carrying out miscalculated instructions prepared in advance, but instead can respond, for example, following a left hand rule as it comes to an opening on the left, whenever the speed and traction of its wheels causes it to arrive there.
In this solution to the maze-solving challenge, a program can manage more than one rule of sense-and-react allowing an LTI robot to respond to various conditions in a maze including light, color, and proximity of walls (using infrared or ultrasonic sensors).However, robots of this design do not learn.In each attempt through any maze, the robot simply transmits information from its sensors through the central program to apply a rule to it as it encounters light, color, or other sensory input much like the proverbial goldfish encountering the landscape new each time it rounds the edge of its bowl.
In order for a robot to not only react, but to gather information and then deploy it in later episodes, the central program must not merely function as an arbiter of sense-act rules, but also gain and store information through experience and act on it in the future.Here again we would expect to see a central program acting as the robot's "brain" attached to peripheral sensors and actuators.The physical structure could be quite similar to the LTI robot described above: typically a chassis holding the central processor with attached sensors and wheels.Upon first encountering the maze, the robot's program has no memories and so it behaves as an LTI robot, responding to lights, colors, and proximity according to sense-andreact rules in its program.The difference is later, on subsequent runs of the maze in an expanded program that not only selects and applies rules responding to stimuli, but that also remembers what rules were used in previous trials, and with what result.Note that even in this simplest of robot constructions there is no explicit program controlling the robot's behavior.Rather, its behavior results-emerges, if you will, from the combination and configuration of its Cubelet parts.To the extent to which we could say the robot is executing a program, that program is implicit in the construction as a whole.There is no one place in the robot that the program is stored.There is no central processor, no 'brain' to the robot.Behaviors that we deem "intelligent" are solely the result of local interactions-there is no representation of the problem or the context, as Brooks (1991) described over twenty years ago in his influential article, Intelligence without Representation.Source: Author

A Cubelets maze-solving robot
Building a Cubelets robot that navigates a maze is surprisingly simple.Figure 3 shows

Figure 2 .
Figure 2.A three-Cublelet 'follow-me' robot made of a Distance and Drive Cubelet.
Figure 3.A Maze Solver: two Distance Cubelets, two Drive Cubelets and a Battery.

3.1 Cubelets-brief explanation and technical summary
Unlike most other robot construction kits, a Cubelets robot has no single central processing computer or 'brain' module that controls its behavior.Rather, each individual Cubelet acts locally on information it receives from its neighbors.For example, a Distance Cubelet uses an infrared proximity sensor to sense the distance to any object in range and announces that distance in the form of a number (0-255) to each of its neighbors.A Drive Cubelet receives numbers (in the range 0-255) from its neighboring Cubelets and runs its motor at a speed based on these numbers.With individual functions built into each block, the Cubelets in a robot constantly pass numbers from one block to the next-these numbers are generated by Sense Cubelets and are consumed by Action Cubelets.For example, the simplest Cubelet robot that you can build has only one Sense Cubelet, one Action Cubelet, and a Battery Cubelet.Take the three-Cubelet "Follow Me" robot shown in figure2.The Distance Cubelet senses when you place your hand in front of it, and produces a number, which it passes to the Distance Cubelet behind it.The Distance Cubelet, oriented to drive "forward" (in the direction of the Sense Cubelet), turns the number into a motor speed, and drives the three-block construction toward your hand.