crossing the river with robots - CiteSeerX

CROSSING THE RIVER WITH ROBOTS: CHANGING THE WAY OF WORKING IN AN AI SUBJECT Eduardo Fermé

Elsa Fernandes

Departamento de Matemática e Engenharias

Departamento de Matemática e Engenharias

Universidade da Madeira

Universidade da Madeira

[email protected]

[email protected]

http://dme.uma.pt/ferme

http://dme.uma.pt/elsa

ABSTRACT In this paper we will describe what we have done in order to change students way of working in an AI subject and we will analyse engineering students’ activity using robots Lego® Mindstroms™ Robotic Invention System™ (RCX and NXT) in AI classes. Keywords Teaching in Artificial Intelligence, developing competences, cooperative and collaborative work.

1. INTRODUCTION Artificial Intelligence (AI) is a field that does not have a generally accepted definition. AI hasn’t got a general theory or unifying principle [7]. AI is the field of computer science that tries to do things that are difficult or impossible to do with traditional programming techniques. From this point of view, teaching AI is an hard task that cannot be approached as a traditional Computer Science or Software Engineering subject. We started teaching AI at the University of Madeira in 2002/2003, and we included, as part of the work, a project, in a traditional manner (i.e.; like a software engineering project). Using project in a traditional manner was a boring task for our students. In general, students began the project ALAP (as late as possible) and their only goal was to be approved in the subject. Under this panorama, we decided to make changes in the kind of work proposed to the students because we wanted to motivate them to learn and to the development of competencies. Our first step was to propose in 2004/2005 the programming of the Othello (Reversi) game, and we proposed a competition among different groups, whose winners would obtain an increase in their final assessment. Our aim when we proposed this task to our students was to promote the interaction between groups. We also wanted students to “do the work in the better way” and not “just do it” as it was usually. The competition between groups and the ‘prize’ was the motivation for work. However, we wanted to reach more than this. We wanted students to cooperate not only collaborate. We wanted them to jointly build solutions for the

problems proposed and to share experiences, partial solutions and mechanical problems. We wanted to retain the “do it in the better way” but we would like that the motivation for work was the challenge proposed by the project itself instead of competition. So we decided to put hands-on and completely change the kind of things proposed to students, because we believe if we want to change students’ way of working we, as teachers, have to modify what we ask of them. In this paper we will describe what we have done in order to change the panorama and we will analyse engineering students’ activity using robots Lego® Mindstroms™ Robotic Invention System™ (RCX and NXT) in AI classes, after the modification.

2. THE DROIDE PROJECT The DROIDE Project: “Robots as mediators of learning” was created in 2005. We placed three kinds of aims for the project: •

to create problems in Mathematics Education/Informatics areas to be solved through robots; • to implement problem solving using robotics in three kinds of classrooms: mathematics classes at K-9 and K-12 levels; Informatics in K-12 levels; Artificial Intelligence, Didactics of Mathematics and Didactics of Computer Science/Informatics subjects at high level; • to analyze students activity during problem solving using robots in this different kinds of classes. Associated to the ‘birth’ of DROIDE project, a laboratory was created where students could work. That brought another dynamic to the classes of the subjects mentioned above, to the kind of work possible to be asked by teachers and to the kind of work developed by students.

3. CHANGING THE WAY OF WORKING Artificial Intelligence is a subject of our degree on Informatics Engineering. So we are interested that our students develop skills, knowledge and behaviours recommended by ABET (Accreditation Board for Engineering and Technology – http://www.abet.org) [1, page 2]. “Engineering

programmes must demonstrate that their students attain: (a) an ability to apply knowledge of mathematics, science, and engineering; (b) an ability to design and conduct experiments, as well as to analyze and interpret data; (c) an ability to design a system, component, or process to meet desired needs within realistic constraints such as economic, environmental, social, political, ethical, health and safety, manufacturability, and sustainability; (d) an ability to function on multidisciplinary teams; (e) an ability to identify, formulate, and solve engineering problems; (f) an understanding of professional and ethical responsibility; (g) an ability to communicate effectively; (h) the broad education necessary to understand the impact of engineering solutions in a global, economic, environmental, and societal context; (i) a recognition of the need for, and an ability to engage in life-long learning; (j) a knowledge of contemporary issues; (k) an ability to use the techniques, skills, and modern engineering tools necessary for engineering practice.” On the other hand AI has its own specificities. In our Informatics Engineering degree the aims of the Artificial Intelligence subject are: to understand what characterizes and distinguishes AI and its fields of application; to know how to represent knowledge and how to reason with it; to use logic (in AI) as a tool for knowledge representation and reasoning; and to learn how to solve problems involving knowledge-based applications. The DROIDE project has its central concern on changing the focus from teacher to student. In that context the AI teacher designed the AI curriculum having, as background, ABET recommendations, AI subject aims and DROIDE concerns. According to that curriculum, an important part of students work was to develop, during the semester, a project where they had to construct and programme a robot to solve a problem using Lego® Mindstroms™ Robotic Invention System™ (RCX and NXT). It is important to note that we were interested in the use of cognitive robots in the course, but this did not mean teaching robotics. The difference resides in the fact that we were interested that students developed knowledge-based programs that would be applied in robots, more than reactive robots. The purpose was creating not fast, but smart robots (for a similar initiative, see [4]).

3.1 Cannibals and Missionaries revisited: For the 2006/2007 academic year, the project proposed to students was a modification of the classical problem of missionaries and cannibals: A group of missionaries and cannibals are in the edge of a river and they want to cross to the other side. The number of missionaries and cannibals are

the same and can be 3, 4 or 5. They have a boat to make the crossing. The boat has capacity of 2 persons in the case of 3 missionaries/cannibals and 3 persons in the cases of 4 or 5 missionaries/cannibals. If during any of the crossings there are more cannibals than missionaries, on one side or on the boat, then the cannibals eat the missionaries. During the crossing the boat shipwrecked and the missionaries sent, to their group, an encrypted message asking help. The message provides us the information about which side of the river the boat is and the total number of missionaries. The objective of the group is to send a robot to finish the crossing. The robot must have the same capacity of the boat and at least one person must be in the boat in each crossing. We can gather the challenges of this project in two different groups: On the one hand, we expected that students solved problems using some techniques of AI, in particular, heuristics search and production systems and related the problem solution with the contents of the subject. On the other hand, in real life, Informatics Engineering problems never come alone, but they include several decisions, challenges, constraints of the environments, etc. that the engineer must mind. Students should face the following constraints: 1) Students had to choose which robot and language they would use. They could choose between RCX and NXT robots. Regarding to the language they could use whatever they wanted. The AI teacher offered support in Prolog for RCX but students had more experience programming in Java. Furthermore, NXT robots are more powerful in terms of hardware resources (more memory, better sensors, four sensor ports instead of three, etc.). But both robots have limitations in terms of motor ports (maxim three) and sensor ports (maxim three or four). They also have inaccuracies concerning motors (two equal motors may have different speeds), performance (the robot performance depends on the state of the batteries) and behaviour (variations in the luminosity of the environment may produce different mensuration from light sensors). 2) The construction of the robot could not be independent of its programming. This constraint avoided that students could divide the project in parts. They had to work together (cooperatively instead of collaboratively) on the construction and programming of the robot. 3) Students could introduce changes in the environment of the project originally proposed

(cannibals were represented by red balls, missionaries by blue balls and river margins were represented by black lines). However, there were many constrains aggregated to the changing: balls were difficult to grab, but where easy to transport (or push), blocks were easier to grab, but more difficult to push. The same shape for cannibals and missionaries helped the transport, but made difficult to distinguish between them, etc. 4) Students could modify the RCX-Prolog Platform. The platform was created assuming three motors, two touch sensors and one light sensor. Using a different robot architecture required the modification of the Big Blocks Prolog programs and the NQC program in the Robot (for an explanation about RCX-Prolog Platform see [2]).

Figure 1 Different Robots Proposed

why of building that robot to solve that problem and of the programming and the environmental options. Having those reports and an anonymous inquiry held to students as our basis for analyses of students work we will try to show emergent aspects of this kind of work for the learning of AI. After carefully analysing students’ reports and inquiries, some patterns emerged. We will go through them in order to share our experience with the reader

4.1 The context of learning According to Lave [5] an arena must be conceived as a relation between person-in-action and settings in relation to which they act. We cannot think about the context as an unique entity but as a relation between arena and setting. Activity is dialectically constituted in the relation with arena. Setting can exist without activity (although it exists because it was created for a certain activity) but arena and activity do not exist one without another, in other words, they only are meaningful in the relation with each other. The fact that there was a laboratory of work specific for this kind of projects obviously promoted a different kind of work, and a different kind of positioning of the students towards the proposed project. Students freely accessed the DROIDE laboratory without the presence of the AI teacher and according to their own availability. Being on the laboratory working with the group colleagues also implied to be with colleagues of other groups who had the same problems to face. These aspects led to the sharing of experiences and ideas as we can read in the below paragraphs written on group students reports1. […] but once proven the inefficiency of this system [referring to a previous robot], we didn’t have another option than to change the [robot] base, and we reused one that had already been done [by other group], which was in a box with a post-it saying “undo if you wish", and that was what we did. (GROUP 3) The experience with the robots allowed us to work with other groups, sharing different knowledge and experiences, which allowed us to overcome some of the difficulties encountered. (GROUP 7)

Figure 2 The same problem, different environments

4. ANALYSING STUDENTS’ LEARNING It was part of the work of the project to write a report that included detailed explanations of the way and

1

Translated from Portuguese to English.

Figure 3 The base of group 10 “reused” by Group 3 and both final versions

Our Project was approached from many viewpoints, but we must recognize that our first meetings were fairly unproductive. We were, meeting after meeting, around a yellow box [the RCX brick] and lots of Lego parts; we had been discussing what to do, how we had to start to form all these stuffs in order to reach the final goal, which was to solve the problem (cannibals and missionaries) with a structure built from this yellow box. It was from this point that first designs emerged: a crane, a mobile arm, a mobile base, all of them common ideas, from the beginning and every new idea we tried to incorporate using the previous designs. (GROUP 3).

Students also shared artefacts. As we can read above (group 3) they shared robots or parts of it that other groups had given up after realizing that construction on that way was not a good idea. They shared programming platforms that were not suggested by the AI teacher and that appeared as consequence of the research done by one of the groups. For all that, final projects presented by students were very different from each other. One positive aspect was the interaction with colleagues who were also doing the project, since it was possible to exchange ideas and obtain valuable information (for example, where to arrange the API for the robot, the interpretation of the error codes, etc.), sparing a long time of search. (GROUP 11)

4.2 Methodology of work adopted by students Students decided what methodology of work they should use to ‘catch’ the work proposed by the AI teacher. In this category we differentiated some aspects, such as the initial approach to the problem, splitting the project in stages, building and programming the robot by trial and error. Students were on a laboratory with thousands of LEGO pieces, two boxes, one yellow and another white – the robot brain - and without any previous experience on using and/or programming robots. They had to face challenges that were posed to them and to acquire elements that allowed them to take decisions. Different groups of students positioned themselves differently towards this initial situation. Some groups took several work sessions only with the yellow box in front and apparently with no idea of what to do or how to catch the problem.

Figure 4 Partial Solutions Other groups started by imagining what kind of robot would be useful to approach the problem (for instance, to build a crane). Originally, we thought of solving the problem by making a robot with a crane. […] But when we tried to implement the game in this way we detected difficulties, not in the construction of the crane, but in the way of recognizing the cannibals/missionaries with the sensor, in the way of transporting them from one margin to the other. In this manner the code would be very complex and their implementation too. (GROUP 2). Initially, we thought of a robot with a swivel tower that carried the pieces, from one side to another, without moving the wheels to make the rotation. (Group 10) Other groups of students opted to catch the problem splitting by it in several sub problems.

1st phase: group meeting. Preliminary planning of the solution, based on the understanding of the problem. 2nd phase: To build up the robot, passing through several adjustments until the final version; to create a scenario, choosing the format of cannibals and missionary; to start the construction of the river and of the banks.

particular putting the crane to rise and fall. (GROUP 11). For the construction of the final version of the robot we had to build many others, which for one reason or another were not appropriate to solve our problem. We built several supports based on the construction of the arm and claw to get the final one. [...] Originally we built a cart to transport the balls from one margin to the other, but this solution was not feasible […] (GROUP 13)

3rd phase: To develop a code in Prolog, to solve the problem. (GROUP 10) Other groups even opted to get acquainted with the robots by following LEGO tutorials.

The group started by building a small robot to discover their potential. Thus, we tested small programs in order to understand what the robot did and to think about our problem and possible solutions. (GROUP 17). We first built up the robot "Robot Educator Model" from the Mindstorms manual taken from the DROIDE laboratory. We felt that it was not the best option because we had to invent everything else (…).(GROUP 12). We started our work by trying to build a robot which could be used later. Therefore we consulted documentation, both online and available in the DROIDE laboratory and we started the construction of two robots at the same time, the Tribot and the RoboArm. (GROUP 11). From all this processes of initial approach to the problem, trial and error emerged as the natural methodology of work. All the construction and programming of the robot was supported by this methodology. So, students built up different robots until they constructed and programmed the final robot, the one that would solve the proposed problem. […] meanwhile we concluded the Tribot, which later proved unsatisfactory to fulfil the task of passing the ball from one side to the other, since their “grab” was too much outgoing and collided the balls when it run. […] We still had the idea of making a grabrobot that moved along the river margins, but it appeared to much complicate, in

Figure 5 First and final versions of Robot Another interesting aspect of this project was the fact that evaluation resides on the student. They did not need the authority of teacher knowledge to legitimate what they had done. They built a robot and they programmed it and, by themselves, they discover what thinks didn’t work. They tried to understand why the robot and/or the programming weren’t working, they undid and re-built. In all these process learning occurs. The first version of the robot had as major problems, the synchronization of the legs to walk; the fact that the ball often jumped from the grab due to the oscillation when the robot walked. This oscillation also gave rise to incorrect readings from the light and ultrasonic sensors. Then, due to this problem, this version of the robot had to be abandoned. (GROUP 17).

Figure 5 First Robot of Group 17

From the methodology of work presented, analysed and discussed above, students easily became aware of the physical and computational limitations of the robot. Originally, we thought that the robot should go in all directions and it meant that we had two engines: one engine to move forward and back and another one to turn. Assuming this, we still needed at least another two engines: one for the arm and one for the grab. However, this was a problem since the NXT has only entry to 3 engines. (GROUP 13) The constraint of using less than three sensors and three motors had caused that our robot would be deprived of some autonomy, for example: by the use of one additional motor we could make the agent allow the discharge of cannibals and missionaries. (GROUP 8). Students thought the constructions with the aim of reducing robots limitations and the margin of error that they had been catching all along the process. After mounted the final robot, it had to undergo some adjustments, in particular, in the number of wheels. Initially we had only used 4 wheels, but we verified that the robot had a misalignment in their way, so we decided to put two additional wheels ahead between the main ones, in order to reduce this misalignment. (GROUP 13). We decided to put the cannibals to the left of the river (in each margin), and the missionaries to the right. So the robot was not to recognize the cannibals and missionaries (the robot already knew that the missionaries were right and the cannibals were left). It just recognized the blue line to collect a missionary or a cannibal, and not needed recognize objects. (GROUP 8)

Figure 6 Version of the river proposed by Group 8 Finally students wrote about what they had been learning with all this work, referring to aspects such as collaborative and cooperative work, researching, decision-making, problem-solving and working in projects. During the development of the project we could test our abilities in several respects: researching, working in groups, problemsolving and decision-making. […] we needed several exchanges of ideas among the elements of the group in order to make the robot as objective and independent as possible. (GROUP 8). In this project our ability to work in groups was improved as well as our skills to solve several problems that were not directly related to the discipline. We applied, in practice, the theoretical concepts of Artificial Intelligence and it helped to understand in which context they are inserted. We noted that some knowledge acquired in previous courses was essential to the project, such as programming in Logic. Despite the difficulties encountered, we found an acceptable solution for the problem. (GROUP 10). After the project was finished, we felt that we could have improved the work done and with a little more time we would have reached that. This project required an elaborated and laborious research on the various topics covered, and we believe that our way of thinking was improved and even our way of implementing future projects. (GROUP 15)

4.3 Students inquiry analyses On the end of the experience teacher held an inquiry, anonymous, to students which aim was to know how students evaluated the experience and to understand if that evaluation go to meet of e teacher expectations. On that inquiry teacher also questioned about the relation between theoretical concepts and the concepts used on the project they had been working. Students identified the following theoretical topics on their practical work: uninformed search strategies, programming with heuristics and they related a robot with an intelligent agent. Students did not become aware that they were working in knowledge engineering, that is, that they were working on the construction of knowledgebased system. According to teacher view this fact is because knowledge representation is an action developed during the project while the concepts identified by students are related with the outcome of the project.

Figure 6 Students working

Students pointed out as main positive aspects of the experience the following: the use of robots to solve an AI problem seeing that AI concepts became meaningful because what they programmed was translated in visible actions; freedom of choice; the sense that robot built belong them (students gave a name to the robot, they talk about ‘my’ robot, and so on); and the discussion held by the group.

From the students activity several elements emerged that belong to the shared repertoire [8] such as artefacts (computers, robots, trays, LEGO pieces, balls and programming platforms) used on working sessions and that had been strongly structuring students practice. In these students activity we could also see styles and ways of ‘doing the things’ that are different from the usual in a traditional class. The fact that all groups worked on the same laboratory, gave each group access to the discussions maintained by other groups and to different ways of thinking that could ‘illuminate’ the problem solution.

Main negative aspects set off by students was that the laboratory was always full of students and robots limitations (memory capacity and sensors limitations).

Another thing that belongs to the shared repertoire was the fact that students recognize impossibility of accomplishing a task without assuming it as their own incapacity.

5. FINAL CONSIDERATIONS Analysing students practice through their reports does not tell us all about their practice (in Wenger sense [8]) but the reports give us enough elements to allow us to recognize the kind of learning carried out by these students. The writing of the report was a learning moment too, not only about writing a report but because students, in order to write it, had to think about what they had been learning and this is not usual when we think about the traditional way of teaching. Students feel comfortable both when building the robots and programming it to solve the proposed problem. Using robots to learn AI promoted an increase of the discussion between students and the collaboration and cooperation on the solution of the proposed problem [3] as well as it increased significantly the students’ level of participation and consequently the ‘quality’ of what they learned.

It was through participation in practice that competence developed. These students had been working at the level of ABET skills, knowledge and behaviours recommendations and at the level of competence development, namely, and according to Niss [6] competence on mathematical thinking, competence on problems management that involves formulation and solution of problems, competence on instruments and accessories that implies to the ability to use and to establish relations with instruments and accessories, competence on communication and competence on representation. We still had evidence that students developed competence on cooperation. With this kind of work students have been becoming more and more independent and autonomous. The way students evaluated their work, increasingly without teacher intervention, showed that they needed the authority of teacher knowledge less and less and that they gained their own knowledge authority. Another important aspect which emerged from our analyses was the way students related and connected the theoretical aspect (studied on the theoretical classes of AI) with their use on the

solution of the proposed problem. It was on practice that the theoretical concepts become meaningful for them. According to Wenger [8], competence is created and defined in action. In other words, it is necessary that the students participate (in Wenger sense) if we want to promote competencies development and this demands time, continuity of work and the students wrapping up with appropriated situations. We can say that the kind of methodology adopted by this AI teacher is a good way of working with students because, labour market wants, more and more, competent workers not only on the area of expertise but it also values transversal competences.

6. ACKNOWLEDGMENTS We want to thanks the referees for their comments. We are also indebted to Mauricio Reis for his help concerning presentation. Eduardo Fermé was sponsored by FCT, POCTI219, Elsa Fernandes was sponsored by the “Centro de Investigação em Educação da FCUL”. FEDER.

7. REFERENCES [1] ABET. Engineering Accreditation Commission. 2006-2007 Criteria for Accrediting Engineering Programs. October 29, 2005 http://www.abet.org

[2] Fermé, E.; Gaspar, L. “RCX+PROLOG: A platform to use Lego Mindstorms™ Robots in Artificial Intelligence courses”. This proceedings. 2007. [3] Fernandes, E., Fermé, E., and Oliveira, R. Using Robots to Learn Function in Math Class. In L. H. Son, N. Sinclair, J. B. Lagrange and C. Hoyles (Eds) Proceedings of the ICMI 17 Study Conference. Hanoi University of Technology. Vietnam. 2006 [4] Kumar, A: "Three years of using robots in an artificial intelligence course: lessons learned". In Journal on Educational Resources in Computing Volume 4, Issue 3. 2004 [5] Lave, J. Cognition in Practice: Mind, mathematics and culture in everyday life. Cambridge. Cambridge University Press. 1988 [6] Niss, M. O projecto dinamarqês KOM e as suas relações com a formação de professores. In M. Borba (eds.) Tendências Internacionais em Formação de Professores de Matemática. Autêntica Editora. São Paulo. Brazil. 2006 [7] Russell, S.J. and P. Norvig, Artificial Intelligence A Modern Approach. Second Edition. Prentice Hall, Englewood Cliffs, New Jersey. 2003 [8] Wenger, E. Communities of Practice: Learning, Meaning and Identity. Cambridge, UK: Cambridge University Press. 1998/