Micro-Robots Based Learning Environments for Continued Education in Small and Medium Enterprises (SMEs).

MARTIAL VIVET [1]

One of the main problems in the Small and Medium Enterprises (SMEs) is the adaptation of their production process to keep their markets. The improvement of the process is often realized by the introduction of new technologies such as more fully automated machinery and computers. Therefore, training has been designed to prepare workers for the future by helping them use these new technologies.

Since 1989, for training people in this context, we have developed learning environments and defined a pedagogical approach, called ATRIUM, which is based on the use of pedagogical micro-robots driven by computers. The micro-robotics activities aim at helping learners discover technology by handling, building, designing, and driving micro-robots through a project-based pedagogical approach. The role of the trainer is to set up and manage the activities of several learner groups (each group is composed of 2-3 persons).

We have experimented with a first generation of learning environments using LOGO as the software to drive the microrobots. Beyond the effectiveness of the approach, we observed that our environments created an overload for the trainers (we say over-appeal) because they had different roles to play. We chose to decrease the trainers' over-appeal and improve their work by increasing the autonomy of the learner groups and by having a better interaction between the computer and the human actors. We have designed and developed pedagogical assistant software, called ROBOTEACH, which cooperates with learners during the micro-robotics activities and helps the trainers manage the sessions. ROBOTEACH is actually diffused and used in professional training in SMEs.

The development of new technologies has been increasing over the past 20 years. We have observed the consequences of training needs: now it is not possible to work without lifelong training. We identify two kinds of lifelong training: the advanced professional training and the reemployment training. Most people are concerned with the latter. Among them, we find the workers in industries who are surpassed by the technology but who must adapt for a new job in their enterprise. The problem with this kind of reemployment training is that you must design pedagogical situations that not only must be adapted to the trainees, but must also be motivating.

For the past 10 years, we have focused our research on the design of learning environments for low qualified adults in the industrial context, taking into account their knowledge and culture regarding technology. We have implemented our environments based on cooperative work within groups, pedagogical projects, and the use of micro-robots.

We begin this article by describing our experience in the design of training situations for low-qualified adults through the presentation of the pedagogical method called ATRIUM, with its activities, tools, and abilities. Then, we present our experiments in industrial contexts. In another part, we explain the limit of our first pedagogical environment. From the proposed solutions, we present the software pedagogical assistant that we have designed, developed, and experimented. Finally, we present our current and future research, recommendations, and conclusions.

MICRO-ROBOTS BASED MICRO-WORLDS

In our laboratory, we use various pedagogical supports from simple, modular, micro-robots flexible workshops. We describe, in the next section, the modular micro-robot aspect with the pedagogical aspects as activities, abilities acquired, tools, and learning foundation.

From LOGO Turtle to Robot Kits 

In training, computer-assisted learning tools are often used. At first, a few pupils control real objects. Seymour Papert (1980) is one of the first researchers who has driven a real object, a floor turtle, in the LOGO microworld. Thanks to a few words in LOGO programming language, the pupil can easily control the turtle. Thus, a relationship can be established between a pupil's actions on the computer and the turtle movements.

The floor turtle is a good approach, but it is limited. You can mask the turtle by another object that has the same behaviors, e.g., a car, but you can't transform it into a crane. This is a problem in training industrial workers because you need material that is relevant for the workers and adaptable for different factories. Thus, the turtle is not adapted, but you can use micro-robots which look like scale models of industrial systems. To solve these problems, we use micro-robots built from Fischertechnik[c] robot kits. The kits are composed of various bricks (motor bricks, captors, assembling bricks, etc.), and allow the building of very different micro-robots (see Figure 1). We activate these micro-robots with a computer.

Another interesting facet of these modular micro-robots is that the trainers can create many learning situations in different contexts (e.g., mathematics; Vivet, 1989a), technology, and programming. It's the reason we speak of pedagogical micro-robotics. In the next section, the main activities in the technology context are presented.

Activities With Modular Micro-Robots

Our pedagogical goal is not to create specific learning tools to use with professional equipment. We prefer design environments for discovering basic technological cultures (e.g., what is a motor, a captor, kinematics chain?) to understand the functioning of automated machinery. Our approach is centered on problem-solving tasks to be achieved by the learners. To accomplish this pedagogical goal we have defined four main activities, which are:

* Driving preassembled micro-robots: the learners use programmed commands given by the trainer to solve problems such as driving a micro-robot in a labyrinth.

* Building a micro-robot from a handbook: once the micro-robot is built, the learners must drive and program the movement for the device.

* Project: the learners must design, build, and drive a micro-robot from a technical directive notebook. A project is defined by its pedagogical goal, technical goal, content, and constraints (e.g., costs, time, viability, maintenance, evolution, security) (Vivet, 1992). The project is a very important point of our pedagogical approach. We manage it as industrial projects. For the beginning of the experiments, we have a specific project that is a design for a machine to sort coins. More than 40 sorters have been designed. Even if the sort system is often the same, there is never the same sorter. The designing and the performance are different. The imagination is not limited.

* Communication and enhancing the production (writing, show, discussion, argumentation about the micro-robots): written and oral communication is essential. An experiment is finished only when the learners can discuss it with other learners. This sharing is only possible by written and oral communication. We allow this sharing in the reception session. During the reception, the learners give a report on the micro-robot and present their productions in front of all the groups of learners and a jury, which is external to the training. We try to evaluate the learners during the discussion. This is a good time for the learners to return to what has been worked on, discovering solutions by others, justifying one's own solutions, even accepting that the others' solutions are better.

A common characteristic of these activities is that they are organized for a group of two or three learners. It is very important because the members of a group must communicate to complete an activity. The projects and the group work are the basis of our pedagogical approach. When we speak about the interaction between learners and a system, we will always speak in the context of a work group.

Abilities

Of interest in the various activities that we defined is they allow the acquisition of a multitude of abilities. We distinguish one more in particular: the resolution of a problem. We highlight it because it is common to all the activities suggested, even when the degree of complexity of the problem varies in function, the material used, or the activity itself.

With the building of micro-robots, the learners can acquire technical abilities such as the understanding of technical diagrams, the precision in the application of a assembly plan, the diagnosis of breakdowns, and maintenance. The learner can also acquire the organization of the space and time, localization in space, and dexterity, which are more psychomotor abilities. Imagination, invention, and inductive reasoning are abilities that are not easy to plan in traditional training settings. The project allows it, as well as the linguistics abilities (writing reports, oral presentation).

As soon as it is a question of micro-robot control, the knowledge identified, e.g., the creation of programs, anticipation, organization, and the planning of a series of events, are implicitly approached. Indeed, the learners are in a setting where they handle real objects. They also understand how the objects function, and they want to drive them (the play aspect of the objects reinforces their motivation). They are consequently obliged to write correct programs to succeed. The use of real objects such as the micro-robots is very well adapted for a programming initiation (Leroux, Vivet, & Brezillon, 1996a).

The management of social interactions in a group is not an area that we examined, but it is intrinsic to the pedagogical method because all the activities are made in groups of two or three persons. It is necessary for each one to collaborate with others to design something. The motivation is the resolution of the problem, so communication with others is natural. Thus, it is easier to work and correct it.

The handling of models (Vivet, 1989b) is an interesting skill which, in our view, is more easily accessible in our training context with real objects. We want to develop abilities such as:

* Finding the relations (resemblance, difference) between the real world and the model of the real world, making predictions based on these relations.

* Being in uncertainty and to explore from this uncertainty.

The ATRIUM Approach

The pedagogical method ATRIUM consists of defining in a situation, the abilities to acquire the associated activities and choosing tools which support the activities. We synthesize the method in the form of the diagram shown in Figure 2. The first axis describes the various abilities possible to acquire in our micro-robotics contexts. The second axis specifies the activities according to the skills defined on the preceding axis. The last axis gives the list of the tools to be used according to the activities.

Due to the links between the axes (links represented by features on the axes), it is easier to define the training situation as a mapping, linking items from the three axes. For example, if the goal is the understanding of technical diagrams, we will set up an activity of building a modular micro-robot from a handbook. If the goal is to develop imagination, the trainer can prepare a special project.

The ATRIUM method defines a general framework for training. The diagram of Figure 2 is not exhaustive, but represents the basis of the method. The content of each axis can vary according to the trainer, the available material, the learners (pupils, low qualified adults), the context of training (traditional teaching or professional training), etc.

Learning and Pedagogical Foundations

The ATRIUM pedagogical method is based on Piaget' s and Vygotski's learning theories and micro-world approach (Papert, 1980). The training process in this method is not limited to a confrontation from the isolated learner with reality, but includes a collaboration with its environment and especially with others (Vygotski, 1962). Learners must be placed in a context where, using materials (driving software, modular micro-robots), they build and acquire their own knowledge. It is essential to create the context which allows acquisition of mental models, these contexts take the form of "objects to think with" (e.g. micro-robots).

EXPERIMENTS IN INDUSTRIAL CONTEXT

For adults, we focused the experiments with low-qualified persons in two projects: QUADRATURE (Vivet, Bruneau, & Parmentier, 1990; Vivet & Parmentier, 1991) and PLUME (Parmentier, Morandeau, & Vivet, 1993). We describe these experimental training projects whose content is oriented towards Computer Integrated Manufacturing (CIM). At the early stage, we used only LOGO as software to drive the micro-robots. It is an important point to understand the following sections about the limits of our pedagogical environments.

The QUADRATURE Project

The first experiment proceeded within the framework of the QUADRATURE project, financed by the Ministry for Research and Technology. This project lasted two years (from December 1989 to October 1991) involving an important car building subcontractor (1,000 employees).

The learning context. The objective of this experiment was to set up, with our tools, training sessions based on the acquisition of basic competencies for workers who should control flexible, automated lines of production in an industrial company. The trainees (two groups of eight workmen) were low-qualified adults who work on the tailored production lines. The goal for the company was to train the operators with new technologies in production without losing the knowledge acquired by these operators over time. It should be noted that this training organization is always used within the enterprise.

The training is composed of three teaching workshops: micro-robotics activities, technical unit (specialization in pneumatic, hydraulic, industrial automates, etc.), and internal training (safety, quality, etc.). The teaching workshops were not contiguous in time to avoid tedium and to gradually bring trainees to a sufficient technical level to control the flexible, automated lines.

The first part of the micro-robotics unit (4-5 days) consists of a progressive catch in the hand of the equipment and the software through activities such as building and driving an arm (Figure 1). The second (5-7 days) project is devoted to the design of a coin sorter (Figure 1), with presentations of the productions in front of a jury.

Experiment results. This project indicated an interest in using the ATRIUM approach before specific professional training for low-qualified adults. Thanks to the modular micro-robots, the trainees could mobilize their knowledge and skills acquired after 20 years of production by designing devices that integrate techniques observed in the factory. The pedagogical strategy is based on the professional life of the workers in order to better structure their knowledge. After this step, we assume that it is easier for workers to acquire new knowledge and skills during the other training activities focused more on the industrial world (flexible manufacturing system and arm manipulators). Another beneficial effect of ATRIUM was observed on the motivation of the trainees. Their participation was active throughout the training sessions.

PLUME Project

The PLUME project took place from November 1991 to April 1993 in a small family business (80 employees) specializing in the production of religious medallions, pins, and key rings.

The learning context. The training was mainly based on the acquisition of basic competence in reading (studies of text, reading diagrams, and schemata), writing (vocabulary, grammar), arithmetic, and problem solving. The training was divided into three workshops: communication, micro-robotics, and CAL. The micro-robotics activities are the same as the QUADRATURE project. The trainees (eight women and seven men) were low-skilled, manual labor workers. They were volunteers taking part in the training. Their education was only at a basic, secondary level.

The training must not disturb production too much. Also, the company is 90 kilometers from the towns where the trainers are based. For these reasons, the training is a combination of distance and local training sessions. The local sessions take place every two weeks. During them, all the trainers and trainees are present in the firm's premises. Each week the trainees have two hours to work on their individual training plan or to continue their activities in a group. They work in the factory, and they can ask trainers for help through a tele-assistance system.

For the micro-robotics activities, the tele-assistance system is based on two telephone lines: one for the communication between the trainer and the trainees, and a second for the connection between computers (Vivet, Leroux, Hubert, Morandeau, & Parmentier, 1993). The company had no access to more modern networks, resulting in the necessity to use "not too much" "high tech" technology in such contexts to help "acceptability" justified the use of telephone lines. When a group of trainees asked for distance help, the scenario was always the same. The group called the trainer and either had direct contact for synchronous communication, or left a message on an automatic answering machine for asynchronous communication. In the later case, the trainer studies the received messages and gives some feedback.

At this point, two situations can arise: either the help given by telephone is sufficient, or not. If it is sufficient, then the tele-assistance is finished. 0therwise, the trainer takes direct remote control and intervenes on the computer of the trainees, thanks to the second telephone line and a specific software. Thus, the trainer can activate the trainees' micro-robot or modify the driving programs. The trainer can also ask the trainees to type commands in order to help directly.

Experiment results. For many trainees, the use of a computer was an important event. Few computers have been installed during the training to control the production in the enterprise. The trainees had no difficulty using them. The training contribution on this level was without question.

The micro-robotics workshop became the center of interest during the projects. For example, in the communication workshop, one goal was to prepare reports and presentations for the reception session (Vivet, 1993).

Nevertheless, we have noted a lack of interest in some trainees during the micro-robotics activities. Our approach was based on the personal and/or professional life of the trainees. It is obvious that, with people who do not work directly in an industrial production context and have no interest in technology, the interest is weaker. It was the case for two women in an administrative work setting. As for any training, it is necessary to know the trainees and to define the pedagogical goals. Nevertheless, the interest of the ATRIUM method, used in technology basic training with low-qualified adults, was confirmed in the PLUME project.

COOPERATION BETWEEN HUMAN AND COMPUTER TO DECREASE THE PROBLEM OF TRAINERS' OVER-APPEAL

Beyond the effectiveness of the approach, we observed in the QUADRATURE and PLUME projects that our environments implied an overload for the trainer (we say over-appeal (Leroux, 1992)). In this section, we explain why the trainer is over-appealed and how we improve work and efficiency by increasing the autonomy of the learners and improving interaction between the computer and the trainer.

Over-appeal of the trainer. Although our micro-robotics environments are exciting for the learners, we have encountered problems with beginners in computer science. To activate the micro-robots, the learners must program the driving commands and use a programming language (LOGO). LOGO is sometimes described as a very simple programming language. Nevertheless, it is a real programming language with its own syntax, program editor, and commands. We observed in the QUADRATURE project that the low-qualified adults were having many difficulties using LOGO and creating the driving programs. The programming activities, with its syntactic constraints, were more disturbing than motivating.

We must keep in mind that teaching a programming language is not the main goal of the training. In such a context, the learners acquire two types of knowledge: basic notions in technology and basic notions in programming. It appears better to split difficulties. The introduction of the basic notions in technology would be sufficient in the beginning. Nevertheless, once the learners are accustomed to the environment, it is interesting to take advantage of the micro-robotics environment to teach programming. So, a new research challenge appeared: to design and develop software tools useful enough to help learners write driving programs and activate their micro-robots.

To limit the disturbance of the programming activities, the trainer must frequently intervene to give explanations about programming language and programming environments. When all groups program their micro-robots at the same time, the trainer is very busy. Sometimes, communication with the learners becomes too important. The traner is overloaded by an over-appeal from the learners. This over-appeal has an effect on the training: it is less effective. To decrease the trainer's over-appeal and achieve the pedagogical goals in the QUADRATURE project, we introduced two trainers for nine learners. These training experiments were very expensive. So, large scale dissemination wasn't possible without adapting our pedagogical environments for training with 15 learners and one trainer.

Before designing a new pedagogical environment with programming help tools, we used the PLUME project to better observe the pedagogical situation. Thus, we defined more precisely the needs and, therefore, the characteristics of the next micro-robotics environments.

Thanks to the PLUME experiments, we found that the over-appeal of the trainer is not the only result of programming problems, even if they are very important. The trainer is overloaded because of the different roles to play:

* To transmit information, knowledge, and exercises with the help of reading sheets.

* To teach programming concepts to help the learners program the movements of their micro-robots.

* To help the learners activate micro-robots and debug command programs.

* To manage activities between the different groups of learners, and manage their pedagogical interventions during them.

These results show that the introduction of learning software in such contexts is not so easy. Its use has an effect on the pedagogical situation that is not always foreseeable. In our case, the trainer must play a role of mediator between the learners and the learning software. Numerous trainer interventions imply a difficult management of the different groups of learners. The over-appeal of the trainer is mainly the consequence of the lack of learners' autonomy with the learning tools, including the learning software.

We have observed the problem of over-appeal in the particular context of training with micro-robots. Nevertheless, it seems to us that this problem is not limited to the micro-robotics environments. We think that the over-appeal appears generally in micro-worlds when the learners lack sufficient autonomy with the learning software and when the trainer cannot manage the progress of the activities. The improvement of the training depends on two kinds of help: one, for the learners during the activities and another, for the trainer to manage the activities. In fact, they are linked because when the system helps the learners, it plays the role of the trainer and helps the trainer in the management of the activities.

We think that the solution to decrease the over-appeal depends on: (a) the researchers' abilities to take into account the trainer in the design and development of the learning software, (Vivet, 1990) and (b) the integration of Artificial Intelligence techniques in the design of help tools.

Increasing Autonomy of the Learners

The majority of trainer's appeals appeared during the activities to solve the problems of the learners. We think that two solutions are interesting to reduce this kind of appeal: either the system helps the learners when a problem appears, or the system cooperates with the learners from the beginning of the activity.

In the first case, an approach to implement is using Artificial Intelligence techniques as the expert systems in the learning units, or directly programming the advisors in the software. For example, in our micro-robotics contexts, these techniques are used to help learners solve the technical and electrical problems. When a learner begins handling the material, it is not possible to know all the possible technical difficulties. The best way to understand that is to build a micro-robot and drive it. The beginning of the driving step is very important because this is the moment where a learner discovers the mechanical and electrical problems. The learners can solve most of them, but some may be not easy. We worked for a long time with the material, and we now have the expertise to debug the material problems. We implemented the expertise in a knowledge basis of an expert system. In our case, we preferred helping the learners to diagnose the problem and discover the solution rather than give the solution direc tly. Thus, the expert system can support the learners from the specified problem.

Research in human-machine cooperation focuses on a joint problem-solving process by the human and the machine (Fischer, 1990; Woods, Roth, & Benett, 1990). Such "joint cognitive systems" aim to accomplish tasks that neither the human nor the machine may do alone. Thus, cooperation progresses by each helping the other. This approach seems very interesting to increase the autonomy of the learners. The design and development of such computer-based environments are based on an analysis of the tasks made by the system or the learners. We have already explained that the learners had many difficulties creating driving programs in our micro-robotics contexts. It is this reason why we have defined a cooperation space and tools to facilitate it between the learners and the system to create the driving programs together.

Indeed the learner group knows the micro-robot because it built and handled it. Moreover, it knows what movements the micro-robot must do. However, the learners have difficulty writing driving programs. Conversely, the system can embed knowledge to automatically generate driving programs. But the generation depends on the structure (kinematics chain of the axes) and the components (motors, captors) of the micro-robot. For that, the system must have information about the device. So, the learners and the system can cooperate to create the driving programs. The learners describe the micro-robot to the system and the system generates the driving programs from the description (more details are available in the section "Cooperation Between ROBOTEACH and Human Actors During One Activity of the Project").

The system can directly increase the autonomy of the learners in their tasks, but it can increase that indirectly, for example, by supporting the learning activities. We explain this point more precisely the next section.

Improving Interaction Between the Computer and the Trainer

We often try to develop complex systems to help the trainers in the animation of their pedagogical sessions. During the QUADRATURE and PLUME projects, one reason for trainer over-appeal was the delivery of information, knowledge, and exercises with the help of reading sheets. In such a case, the solution to reduce the over-appeal is simple: the documents describing the activities must be implemented in the system, and this presents the activities to the learners following a plan defined by the trainer before the session. The autonomy capacity offered to the learners depends on the quality of the system architecture designed. The more the trainer can easily plan and set parameters of the activities, the less interventions there will be during the sessions to manage the activities. In this case, the richness of the computer solution comes from the architecture of the system rather than its intelligence.

As the cooperation between the learners and the system increases their autonomy, we think that the partnership between the trainer and the system must improve the management of the activities. This cooperation can take place at three different moments: before, during, and after the session. Before the session, the trainer and the system cooperate to plan the activities. During the session, they interact to control the work and to help the learners. After the session, they cooperate to analyze the activities resulting from the events.

For the moment, we reject the complete management of the pedagogical session only by the system. We prefer to set up an interaction between the system and the trainer. The system presents the activities planned by the trainer before the session and supports the learners with its help tools when a problem occurs. The system may ask the trainee and call the trainer when an intervention during the session is planned, when it does not know how to help the learners or has no more available advisors. Thus, the trainer intervenes just in time.

The "just in time" intervention process is very important in pedagogy, but it is not easy to implement. We have begun research about computer-based analysis of event traces. In our context, the analysis allows calling the trainer when the system detects that the learners may be in difficulty. This work seems to be very promising and we give more details in the section on "Analysis of Event Traces."

Here, we have described few generic solutions to increase the autonomy of the learners and to improve the interaction between the trainer and a learning system. In the next section, we present how we have implemented them in a software pedagogical assistant.

SOFTWARE PEDAGOGICAL ASSISTANT COOPERATING WITH A TRAINER AND A GROUP OF LEARNERS

As we explained it in the previous section, we chose to increase the autonomy of the learners, and help the trainer through a real cooperation process between the system and the human actors (learners and trainer). The consequences of this cooperative approach led to a new learning environment model and the development of a software pedagogical assistant (ROBOTEACH) in the micro-robotics context (Leroux, 1995). We describe the model and the application before presenting results about their use in training.

Learning Environment Model Based on Cooperation Spaces

The model of the learning environment (Figure 3) is based on the articulation of two cooperation spaces: (a) a space of global cooperation, and (b) spaces of local cooperation. In the former space, a trainer and the local-cooperation spaces interact. In a local-cooperation space, a group of learners (2-3 persons) and a software pedagogical assistant cooperate, possibly interacting around a device(s).

The devices are not obligatory in our model. The space of local cooperation may be limited to a software pedagogical assistant and a group of learners. Nevertheless, the role of the devices can be important in particular pedagogical situations. For example, in the micro-robotics environments, we use a physical objects-based micro-world composed of various micro-robots, handbooks, and a workshop of various bricks to build devices. The learners design, build, and handle micro-robots from the physical micro-world. The interaction between the learners, the software pedagogical assistant, and the physical micro-world are essential in a micro-robotics context.

The space for cooperation is a central idea in the model. It is a place where cooperation and/or collaboration (intentional and/or emerging) exist at different levels, in different workplaces, and with agents that are heterogeneous by their nature and roles (trainer, software assistant, and learner) (Leroux, Vivet, & Brezillon, 1996a). We use the terms cooperation and collaboration as defined by Roschell & Teasley (1995) and Leroux, Vivet, & Brezilon (1996a). Collaboration concerns a task that is common to all the agents. Cooperation concerns a division of the task into sub-tasks that are treated individually by the agents, with each agent becoming a specialist. Even with differing local goals, the global goal is always the same for all the agents.

In a space of local cooperation, learners collaborate or cooperate with respect to the activities proposed by the software pedagogical assistant. With the elements of the device unit, they must design, build, and handle devices. The role of the software pedagogical assistant is to manage the organization of the activity (the tutoring role) and to optimize the collaboration process among learners by cooperating with them during their activities.

The global-cooperation space contains all the local-cooperation spaces and the trainer. The learner groups may communicate directly with others to provide help, cooperate, collaborate, or compete in concurrent projects.

The trainer is the central agent that authorizes and eventually coordinates or regulates the exchanges between the groups, has an overview of the activities of each local-cooperation space, and may support some learners locally. The trainer may also take appointments with learners for an evaluation purpose. The trainer then cooperates with the software pedagogical assistants that are present in each local-cooperation space to determine and plan the activities of the learner groups.

We have implemented this learning environment model in the micro-robotics context. In this implementation, the software pedagogical assistant is called ROBOTEACH, and we use micro-robots as devices.

ROBOTEACH: A Pedagogical Assistant Software in Technology

ROBOTEACH is a computer-based, interactive learning system designed to train people in control technology. It is designed to improve learning through help to trainers, making them more efficient when they interact with learners. ROBOTEACH cooperates with learners through project-based pedagogical activities. The trainer interacts with it to prepare personalized learning sessions and analyze the micro-robotics activities. Thus, the trainer manages the activities of the different groups of learners in a classroom, and manages the sharing of training time between groups.

It is interesting to note here that, in some sense, we consider the trainer as managing the learner's model (the student model is not in the software but in the teacher's mind) for the strategy function (Self, 1987).

Architecture of ROBOTEACH. For a group, the pedagogical session corresponds to a programmed access to different learning units. This allows basic concepts in technology to be learned, particularly in micro-robot driving.

ROBOTEACH runs in two different modes: the learner mode and the trainer mode. The first mode manages the activities and controls the interactions of the system during the learning sessions. In the second mode, the trainer can plan activities to prepare the sessions and study the traces of the events after the sessions.

ROBOTEACH is composed of three components: (a) the learner resources, (b) the author's tools, and (c) the control units (Figure 4).

Learners resources. The learner's resources represent the supports of the activities in ROBOTEACH. We have organised these resources in five parts: the electronic books of courses, the electronic books of exercises, the description environment, an expert system for on line help, and the driving units.

A. Electronic books of courses and exercises

We have developed a hypermedia that is the basis of the theory. It embeds all notions introduced in our technology context. We have implemented this hypermedia as a network of electronic books (Forty-five electronics books are available for course about technology). Each electronic book introduces one idea, for example, an idea about movement of translation and explanations about the working and use of a motor (Figure 5).

The learners can conduct experiments that give meaning to the theory in the learner's mind. The experiments are made up of driving micro-robots built to explain the idea or activate a restricted assembly built by the learners from a plan (picture on the screen) given in the electronic book (Figure 6). The electronic books of exercises correspond to the exercises created from experiments that we conducted with adults and pupils, and the exercises built by trainers.

B. Description environment and help expert ystems

The description environment allows automatic generation of driving programs from a description of a micro-robot constructed by the learners. The learners have, on the screen, a virtual model of their micro-robot (Figure 7). More details about the interest of this environment are available in the section "Cooperation Between ROBOTEACH and Human Actors During One Activity of the Project."

During the description of a micro-robot, the system can help learners debug their description or check the working of the micro-robot. We have developed an expert system that provides help for this activity. The forms of help are either written messages on the screen or the opening of relevant electronic books.

C. Driving units

The driving units are composed of activating units of specific micro-robots and a programming and driving environment. With the activating units, the learners activate the micro-robots from buttons of the interface. Each activating unit corresponds to a specific micro-robot (Figure 8). The goal of this activity is to conduct a diagnostic check of the micro-robot built from a handbook. In the case of an incorrect operation of the micro-robot, the system can help find the malfunction. After several attempts under the expert system's control, and the problem is not solved, ROBOTEACH requests the trainer to check the problem. The trace of the system reasoning is then available to show the trainer what happened and which trials were made while the trainer was not there.

The creation of programs is made through a specific programming and driving environment (Figure 9). The group of learners creates their own programs by assembling the programs generated by the description environment with algorithmic instructions. Assembling is accomplished by direct manipulation. The algorithmic instructions are often used to drive micro-robots in an environment with obstacles or constraints (e.g., carrying a piece of metal only when the system detects a piece). The programming of the micro-robots allows introducing programming concepts more easily because the learners directly see the effects of their programming activity through the movements of their micro-robots, thanks to ideas coming from LOGO culture.

During the description and programming activities, the learners check their description and programming by direct tests with the micro-robot. The results of their actions are immediate. The execution of movements by the micro-robot reflects to the learner, the description, or the programming. The problems of description or programming are regarded in dysfunction of the micro-robot. We think that the debugging phase is thus facilitated.

D. Authors tools

Two kinds of authoring tools are available: text editor and the generator of electronic books. The text editor allows one to modify the texts of the electronic books or the help screens of the description and programming environments. With the generator of electronic books, the trainer creates and modifies his/her own electronic books of course or exercise.

E. Control units

Two control units manage the running modes of ROBOTEACH. The learner unit manages the learner mode, and the generator of pedagogical session manages the trainer modes. The learner unit manages the presentation of the activities from a plan programmed by the trainer in the generator of the pedagogical sessions. With the generator of pedagogical sessions, the trainer configures the space of local cooperation by defining the physical objects-based micro-world and the parameters of the software pedagogical assistant. The trainer plans the description and programming activities and sets up the software environments, particularly by tuning the degree of help (none, when the learners ask for it, or when an error occurs).

During a session, the learner unit saves events occurring during the session (the actions of the learners, the errors, the schedule of the events) in a text file (we refer to it as the trace of the session). After the session, the trainer accesses the trace, thanks to a specialized unit which presents the trace in a hypertext form in order to facilitate fast access to pertinent information.

Cooperation Between ROBOTEACH and Human Actors During One Activity of the Project

Thanks to ROBOTEACH, the trainer plans the activities described according to the ATRIUM pedagogical method. The project is always the most interesting activity, and it allows a real cooperation between the human actors and the software pedagogical assistant.

Cooperation implies a division of tasks (Figure 10). The learner group describes one micro-robot to the system through the description environment. It must also build complex programs (a complex program is a combination of elementary programs and algorithmic structures) in the programming and driving environment. The description environment generates the elementary programs. An elementary program allows an elementary action of the micro-robot (e.g., activating electromagnet, a translation movement and a rotation movement). The system aims to create the elementary programs and help learners to find the description errors and the micro-robot breakdowns.

When the system has no more knowledge useful for helping learners, it requests help from the trainer. The idea of having systems able to declare themselves "incompetent" is quite important to shift from the Intelligent Tutoring Systems (ITS) paradigm to the software pedagogical assistant/trainer cooperation paradigm. The trainer's contribution to cooperation during the project is intervention in order to get things moving again.

With ROBOTEACH, some projects begin with a description activity and then a programming activity. In the latter, the learners must sometimes modify the description because they have observed errors or they have modified the micro-robot. This is the reason why access to the description environment is possible from the programming and driving environment.

The micro-robot intervenes indirectly in the cooperation. Its role is very important because the learners activate it to check the description and the driving programs. If the result is positive, the description or the program is validated. Otherwise, the learners must modify their work. The micro-robot reflects an image of the description or programs made by the learner group. It reveals the errors of the description, the programming and the driving. The support units for projects, the description environment, and the programming and driving environment have been implemented from a cooperative learning system model based on heterogeneous agents (Leroux, Vivet, & Brezillon, 1996b).

TOWARDS DISTANCE LEARNING AND META-COOPERATIVE SYSTEM

We have tested ROBOTEACH in real training sessions during more than 400 hours with different learners: technical students, pupils of secondary school, and low-qualified adults in industry. These experiments have proven the performance of the system and have shown the interest in using it in different training contexts. Now, ROBOTEACH is more than a research prototype since it is sold in schools and for training in SME. Particularly in the last context, a car builder subcontractor enterprise has used it since 1996 to train workers to become production team members in a CIM production context. More than 150 workers have been trained since the beginning with ROBOTEACH.

Now, our goal is to improve the cooperation between the trainer and the system, particularly in a distance context. In this section, we begin by explaining why we focus our research in this way. Then, we address the analysis of event traces, which appears very interesting in a distance learning context.

New Learning Environments in a Distance Learning Context

With big factories (e.g., in the project QUADRATURE), it is easy to plan training. The problem with the workers in production is minimal because there are enough workers in the company. But in the SMEs, you must adapt the training to the production organization (e.g., in the project PLUME). For the companies, it is often easier to free workers for two or three hours rather than a week, even one day. In this case, the problem is with the trainers. It is difficult for them to come into the enterprise for only a few hours, especially when they live far away. It is this reason why we think that distance learning can allow for flexibility in training in SME in rural areas. The learners work in the enterprise and they can easily communicate at a distance with the trainer. When the numbers of workers is too small and the training is too expensive, it is possible to plan activities with groups of learners from different manufactories. For example, we began designing distance-learning software, which allows cooperati ve projects with learners distributed in different distance places.

The model of learning environments in a distance-learning context is the same as the model described in the section "Learning Environment Model Based on Cooperation Spaces." The difference is in the workplace. The workplace is not in a classroom, but on a network-like Internet. The spaces of local cooperation and the trainer are distributed on the network. Thus, the learning situation is more flexible in space and time. The activities can be synchronous or asynchronous. Instead of a single trainer, several can intervene at different moments. If the numbers of trainers is not important, their cooperation with the software pedagogical assistant will be modified.

When we experimented with ROBOTEACH, we observed that the trainers weren't analyzing the traces of the sessions. The reason is that they are in the classroom and they live the sessions. So, they get a direct representation of the learning situation and its progress. They plan the next sessions by taking into account their own representation. They don't need the traces because they have lived through the sessions. At a distance, when the trainers are not in the workplace, the situation is different. They do not have a direct representation based on which they have been Living. So, the analysis of event traces become very interesting for the trainers to adapt the configuration of the software pedagogical assistant to the situation (e.g., by increasing the degree of system help).

We previously mentioned that the system ROBOTEACH recorded the whole trace of the learner actions. A possible way to use the traces is in establishing a diagnosis of the sessions. Our present ambition is not to evaluate knowledge of learners, but to analyze their observable behaviors (VanLehn, 1988) during a training session. The system must be able to raise, for the trainer, the critical revealed behaviors within the cooperation between a group of learners and a software pedagogical assistant. From there, the trainer will lay out additional information to create and personalize the next sessions.

Wenger (1987) distinguishes, in the diagnosis, the behavioral level and the epistemic level. The behavioral diagnosis is limited to the rebuilding of the actions, which was necessary for the resolution of problems to deduce behaviors without interpreting them. The epistemic diagnosis goes further and consists of an interpretation of behaviors to define a knowledge state model of the learner. In our actual work, we limit ourselves to a behavioral diagnosis.

To solve a problem with ROBOTEACH, the learner group must select some tasks among all those proposed by the system. To achieve its solution, the group interacts with the system and creates its own track within the tree of tasks. After analyzing traces of learner groups, we have observed that the trace along the way is sometimes very disrupted. We speak of fast navigation (Despres & Leroux, 1997). During the session we have marked the moments where the learners had problems. In fact, these moments correspond to disruption time of the traces.

From these observations, we have designed and developed a diagnosis system able to detect the disruption of the traces and alert the trainer. The system is based on an extractor of facts and an expert system which analyzes the facts to produce the behaviors of the learner group (Despres & Leroux, 1997).

In the current state of our research, the session diagnosis is limited to some behaviors. A finer diagnosis seems possible if we analyze the whole interactions, particularly those at a lower level. Such a diagnosis seems very promising within the context of design of distance learning.

CONCLUSION

We have presented our research on intelligent systems in a lifelong learning context. We have described modeling situations of computer-based learning and the interesting results in real professional training contexts.

From these experiments, we have defined a new model of a learning environment based on cooperation spaces where software pedagogical assistants cooperate with human actors. Our goal is to decrease the over-appeal of the trainer by increasing the learner autonomy and improving the interaction between the system and the trainer.

We have cemented our ideas by the implementation of a software pedagogical assistant called ROBOTEACH. The system has been experimented with different types of learners (pupils, technical students, and low-qualified adults in SMEs). Now, ROBOTEACH is used in classroom and professional training.

We took advantage of this long experience in the design and development of computer-based learning environments to focus our research in new directions, such as distance learning. The goal is to facilitate training organization. This challenge is important, but we think that future applications will be based on these approaches.

Note

(1.) Martial Vivet passed away on Thursday October 21,1999.

In 1994, he arranged a working conference in Nancy on behalf of 3.3 and 3.6: "The Virtual Campus: Trends for higher Education and Training." Martial was involved in the detailed planning, not only of the technical programme, but also in the organisation of the conference. Again, he was a wonderful host, and all who attended that conference will remember it for the wonderful social occasion as well as a great technical success. In addition to the technical work he did for WG3.6, we will also remember him for his personal human qualities. He was always very kind to everyone he met, showing great sympathy and understanding to all. We will always think of him with fond memories.

Sent to the many colleagues of Martial Vivet, by Jan Wibe and Gordon Davies; Chairmen of the IFIP WG3.6, TC3.

Thu, 28 Oct 1999 09:32:05 +0200

References

Despres, C., & Leroux, P. (1997). Raisonner sur la trace: Analyse de sessions avec l'application ROBOTEACH. In Actes des 5emes journees EIAO de Cachan. Paris, Hermes, 277-288.

Fischer, G. (1990). Communication requirements for cooperative problem solving systems. Inform actions Systems, 15(1), 21-36.

Leroux, P. (1992). A new development of control technology. Control technology in elementary education. NATO ASI Series, series F, 116: 241-252. Berlin: Springer Verlag.

Leroux, P. (1995). Conception et realisation d'un systeme cooperatif d'apprentissage - Etude d'une double cooperation: Maitre/ordinateur et ordinateur/groupe d'apprenants. Doctoral dissertation University of Paris 6.

Leroux, P., Vivet, M., & Brezillon, P. (1996a). Cooperation between humans and a pedagogical assistant in a learning environment. Proceedings of the Second International Conference on the Design of Cooperative Systems (COOP'96). Juan-les-Pins: France, 391-404.

Leroux, P., Vivet, M., & Brezillon, P. (1996b). Cooperation between a pedagogical assistant, a group of learners and a teacher. Proceedings of European Conference on AI in Education (EuroAIED). Lisbon, Portugal, 379-385.

Papert, S. (1980). Mindstorms: Children, computers and powerful ideas. New York: Basic Books.

Parmentier, C., Morandeau, J., & Vivet, M. (1993). Recherche PLUME, Rapport de recherche pour le Ministere de Ia Recherche et de l'Enseignement Superieur. Nantes, France, Cahiers du CARIF - Hors serie.

Roschelle, J., & Teasley, S.D. (1995). The construction of shared knowledge in collaborative problem solving. Computer Supported Collaborative Learning. Berlin: Springer Verlag.

Self, J.A. (1987). Students models: What use are they ? Artificial Intelligence Tools in Education. IFIP TC3 Working Conference, North-Holland.

VanLehn, K (1988). Student modeling. In M.C. Poison & J.J. Richardson (Eds.), Foundation of Intelligent Tutoring Systems. LEA, 55-78.

Vivet, M. (1989a). Micro-robots as a source of motivation for geometry. Intelligent Learning Environments: The Case of Geometry. Berlin: Springer-Verlag, NATO ASI Series F, 117.

Vivet, M. (1989b). Knowledge based tutors: Towards the design of a shell. International Journal for Educational Research (IJER), 12(8): 839-850.

Vivet, M. (1990). Uses of ITS: Which role for the teacher. New directions for intelligent tutoring systems. Berlin: Springer-Verlag, NATO ASI series F, 91, 171-182.

Vivet, M., Bruneau, J., & Parmentier, C. (1990). Learning with micro-robotics activities. Integrating advanced technology into education. Berlin: Springer Verlag, NATO ASI, Series F: 139-148.

Vivet, M. (1992). Educational uses of control technology. Control technology in elementary education. Berlin: Springer Verlag, NATO ASI Series, series F, 116, 173-182.

Vivet, M., & Parmentier, C. (1991). Low qualified adults in a computer integrated enterprise: An example of inservice training. Proceedings of IFIP /TC3 meeting (Alesund, Norway). North Holland: 261-272.

Vivet, M. (1993). La reception des productions des apprenants: Une phase essentielle dans Ia conduite de projets. Actes du quatrieme colloque international sur la robotique pedagogique. Paris: INRP, Lieges, University of Lieges: 117-134.

Vivet, M., Leroux, P., Hubert, O., Morandeau, J., & Parmentier C. (1993). Teleassistance of trainees in an SME: A case study. Teleteaching (A29). Elsevier Science Publishers B.V. (North-Holland), 907-914.

Vygotski, L. S. (1962). Thought and language. Cambridge: MIT Press.

Wenger, E. (1987). Artificial intelligence and tutoring systems. Computional and cognitive approaches to the communication of knowledge. Los Altos: Morgan Kaufmann.

Woods, D.D., Roth E.M., & Bennett K. (1990). Explorations in joint human-machine cognitive systems. In S. Robettson, W. Zachary, & J.B. Black (Eds.), Cognition, computing and cooperation. Norwood, NJ: Ablex, 123-158.