From: AAAI-87 Proceedings. Copyright ©1987, AAAI (www.aaai.org). All rights reserved.

AN INTELLIGENT TUTORING SYSTEM FOR INTERPRETING GROUND TRACKS

DT. Kathleen Swigger* Lt. Cal. Hugh Burns Harry Loveland Capt. Terresa Jackson

Air Force Human Resources Laboratory Intelligent Systems Branch Brooks Air Force Base, Brooks Texas 78235 Abstract loops, direction, etc., and then use this information to This paper describes an intelligent tutor- “estimate” the orbital elements. In order to duplicate ing system for the space domain. The system this nrocess. we decided to build a qualitative model of was developed on a Xerox 1108 using LOOPS how *the exbert nredicts orbital elements, and then use and provides an environment for discovering this model Awitgn a microworld, or simulated environ- principles of ground tracks as a direct function ment, that allows the student to manipulate various of the orbital elements. The system was orbital elements and observe how each of the parame- designed to teach students how to “deduce” a ters affects the shape of the ground track. satellite’s orbital elements by looking at a graphic display of a satellite’s ground track. B. Student/Computer Interaction The system also teaches students how to use As nreviouslv mentioned. the microworld for the more systematic behaviors to explore this Ground Track problem offers a number of online tools domain. Since the system is equipped with a that permit students to discover relationships between number of online tools that were specially orbital parameters and ground tracks. This environ- designed to help students better understand ment consists of an elaborate ground track display (Fig- facts, principles and relationships, the student ure 1) and a number of interactive tools designed to is free to investigate different options and encourage systematic behaviors for investigating-ground learn at his own pace. track related problems. The student initiates a I. Introduction discovery activity by changing one or more orbital A. General Introduction parameters or changing the injection parameters. This One of the nine basic operational missions for the Air task is accomnlished bv nositionine: the cursor over the Force is the continuous monitoring of the exoatmos- individual Da&meters and messin; the left mouse but- pheric arena through ground and space surveillance. ton to in&ease the value& or th\ middle button to NORAD, through its Space Defense Center, maintains decrease the value. The injection point is changed by a worldwide network that senses, tracks, and analyzes positioning the cursor over a particular point on the map and pressing the left mouse button, which the characteristics of orbiting systems. automatically sets both the longitude and latitude. A In order to monitor and plan for satellite missions, student can- observe the results of these changes by the Air Force crew must be able to read and under- selecting Generate a Ground Trace from the- main stand ground tracks. Ground tracks are two- menu. After investigating the effects of changing dif- dimensional displays that show the portion of the earth ferent narameter values for different ground tracks. the that a satellite covers in one orbit. If you can imagine student can advance to the Prediction window where being placed inside a satellite and being able to look he can make a hypothesis regarding the particular directly down on the earth, then the “ground track” is shape of a ground track. that portion of the earth that you would see as you travelled through space. The ground track is a direct In the Prediction portion of the program, the sys- function of the orbital elements, so proper understand- tem displays a list of words that describe various ing of these functions and of the interactions between features -about ground tracks such as shape, size, and orbital elements is critical for anyone interested in symmetry Figure 2). From this list of descriptors, the satellite operations. student seI ects the words that “best” describe the current ground track under discussion. The student One way to teach students how to deduce orbital then tes& his prediction bv selecting this option from elements from a satellite’s ground track is to present the menu and* comnarine: -his innuts to the Expert’s the various mathematical formulas that are used to conclusions. The &deGt can then interrogate the compute the orbital elements and then show how to Expert System by placing the cursor over aflvy of the apply these formulas to situation- specific tracks [Bates descriptors and pressing the left mouse button. A et al., 1971; Astronautics, 19851. In contrast to this “Why” pop-up menu appears on the screen which the approach, we discovered that experts store ground student can mouse and receive an exulanation of the tracks as graphical representations, indexed by feature expert’s reason for the correct descripior. The student and shape. Based on previous experience, experts learn can continue this iterative process of changing parame- how to detect specific features such as size, number of ters, making predictions, and asking why until he understand the various relationships between After *The research reported herein was supported, in part, with funds making several successful predictions, the student from the Air Force Office of Scientific Research (AFOSR), and enters-a Test environment which is designed to check sponsored by the Air Force Human Resources Laboratory, Brooks the student’s predictive powers by &king him to AFB, Texas.

72 Al & Education perform a task in the reverse order of the one described Model, and (6) .the Coach. The Expert Module includes above. The student is shown a specific type of ground the rules and inference procedures used to deduce track and asked to enter a “guess-estimate” of the shape descriptors from a set of orbital parameters. The corresponding orbital parameters. If the student is suc- Curriculum Module contains the major concepts associ- cessful, then he can continue to explore different types ated with the ground track domain. The State Module of ground tracks. If the student is unsuccessful, then he contains a list of appropriate behaviors for exploring receives information about why his answers are the microworld. The Diagnostician is a set of software incorrect. procedures which evaluates the student’s answer, analyzes student errors, and updates the Student and C. Tool Description Curriculum Models. The Student Model stores the student’s current state of knowledge of both ground There are three major online tools that can be used by tracks and effective tool use. The Coach contains the the student to gather information and to understand instructional rules that tell the system when to inter- concepts and principles about ground tracks. These vene. The Coach makes this decision based on informa- tools are a) a History Tool that allows the students to tion it receives from the Student, Curriculum, and overlay previously generated ground tracks and note ‘State Models regarding the student’s current state of relationships between parameters b) an Orbit Window knowledge. A more detailed description of each module that displays a two-dimensional representation of the is presented below. orbit (Figure 1); and c) a Definition/Example tool which displays factual information about different orbi- B. The Expert Module tal parameters (Figure 1). This Module contains the rules and procedures used to The History tool is specifically designed to help deduce shape descriptors (e.g., closed-body, symmetri- students recognize relevant patterns between and cal, vertical; compressed, lean-right, hinge-symmetry, among previously generated ground tracks. As the stu- with loops) from a set of orbital parameters (eccentri- dent generates various ground tracks, the system col- city, period, semi-major axis, argument of periapsis, lects and stores each transaction. The student can inclination). The Expert Module is invoked only when retrieve any of thii data by selecting the History option the student is making a prediction or is in the Testing from the main menu. A lit of the past twenty ground mode. The Expert Module works by posting a series of tracks appears on the screen from which the student goals which determine the various shape descriptors. can select one or more related ground tracks. The sys- The general problem solving strategy employed by the tem then overlays the selected ground tracks onto a Expert Module is to determine a shape descriptor by single map. Again, the student observes the results of examining a specific orbital element. If this fails, then this exercise. the system looks at another shape descriptor and For any given set of orbital parameters, the stu- attempts to find its value, or looks at a combination of dent can obtain a two-dimensional display which shows two or more orbital elements to see if the system can the position of the satellite in relationship to the earth. deduce a shape descriptor. For example, the Expert The student selects the option labelled Orbit Window Module determines the symmetry shape goal by asking and gains immediate access to this particular display. whether this is a circular orbit. If the orbit is classified The Orbit Window is especially useful for demonstrat- as a circular orbit, then its eccentricity must be equal ing the relationship between the ground track and the to zero. If the orbit is elliptical then its eccentricity is actual orbit and for illustrating the effect of perigee on not equal to zero and the Expert Module must look at elliptical orbits. the orientation descriptor, which in turn must look at The Definition/Example tool provides the student the argument of periapsis. In thii manner, the Expert with the factual knowledge about various parameters. Module can determine a set of shape descriptors for a A student can obtain definitions and examples for both given set of ‘orbital parameters (and vice versa). During the orbital parameters and the shape descriptors by the process of deducing shape descriptors, the Expert simply placing the cursor over the keyword in question Module also determines the optimal “procedure” for and pressing the right mouse button. A pop-up menu deriving the shape descriptors. Thus both declarative appears on the screen from which the student can and procedural knowledge is available to the rest of the select either the definition or example. tutor. Thus by using the available tools, a student can Another function of the Expert Module is to obtain facts about the orbital world (through the deduce parameter descriptors (such as a Circular, Syn- Definition/Example tool), see relationships between dif- chronous orbit) at the same time that the system is ferent ground tracks (through the History window), deducing the shape descriptors. These parameter descriptors are used by the Curriculum Module to and understand certain principles about satellite opera- determine the essential skills that are necessary to tions (through the Orbit Window). A student has the understand a given ground track. Since the rules for option of using any of these tools at any time during determining the Curriculum Skills are embedded within the computer/student interaction. If, however, the stu- the Expert Module rules, we now describe the organiza- dent is not making sufficient progress, the system inter- tion of the Curriculum Module. rupts and directs the student to use a specific tool to achieve an objective. 6. The Curricnlum Module II. Design of the System Along with knowledge about shape descriptors for ground tracks, a student must also understand how A. Overview thii information relates to specific orbit types. For example, an orbit which has a semi- major axis equal to The system is composed of six major parts: (1) The 42,250 kilometers is said to be in a synchronous orbit. Expert Module, (2) the Curriculum Model, (3) the This term applies to all ground tracks that have a State Model, (4) the Diagnostician, (5) the Student semi-major axis equal to 42,250 kilometers, regardless of

74 Al & Education the numbers that might appear for the other orbital cian looks at the Expert Module and obtains a list of parameters. Thus it is important that students recog- the orbital elements which were used to make a correct nize the relationship between the specific domain prediction. The Diagnostician then looks at the the knowledge and the qualitative model produced by the student’s History file to see if the student is manipulat- Expert Module. The Curriculum Module, therefore, ing the correct parameters. If not, then the Diagnosti- contains the specific content that is used to categorize cian invokes some high- level rules that try to generate different orbit types. This knowledge is stored in the the student’s error to match the student’s input. Some Curriculum Module according to how it is used (and of these high-order rules are: Look at the rules that are deduced) by the Expert Module. For example, the used to deduce this shape-type and drop all the AND Expert System determines whether an orbit is circular portion of the rules and change them to QR9s. (Student or elliptical as it deduces the symmetry goal. The Hug: An overgeneralization of a rule . Look at all the knowledge about shapes and orbit types are part of the rules that deduce this shape-type an cl find the “easiest * Expert System. rules (i.e., rules with one or two constraints) and see if The Expert Module also provides a very powerful this is the parameter that the student is manipulating tool for organizing the content areas and for determin- (Student-Bug : If a rule works in one case, it works in ing various levels of difficulty. For example, the rules all cases). that determine the shape descriptors associated with The Diagnostician also monitors the student’s use circular orbits tend to have fewer constraints attached of the various tools. Every time the student selects a to them, and also tend to be fired first, and, as a result, different activity, this information is passed to the Stu- tend to be easier for the student to learn. The hierar- dent Module. chy of orbit types as represented in the Curriculum Module shows both the order that the knowledge should be learned and the relationships between the The Student Model contains a record of the student’s knowledge. This information is used by the Coach to current understanding of both the domain knowledge recommend easier problems whenever the student and investigative behaviors. Whenever the student becomes confused. tests a prediction or changes parameters in the Testing Mode, the Diagnostician sends the Student Module a . list of the rules that the student understands. The The State Module contains a list of goals and subgoals Student Model maintains a series of counters for each which presumably indicate acceptable procedures for rule indicating the number of times a rule is used exploring the Microworld. As the student proceeds appropriately, inappropriately, or ignored (a “missed- through each of the states, the tutor records his/her opportunity” as defined in Carr and Goldstein, 1977). If actions. The authors have hypothesized that a student the missed-opportunity counter exceeds the used- indicates appropriate experimental behaviors if they, appropriate counter, then the Coach recommends first, explore the microworld. The student explores a intervention. microworld by generating ground traces. The student The system also records the number of times that then moves on to “making predictions,” followed by an online tool is invoked. In addition to this counter, testing and validating tests, and then generalizing these an effectiveness measure is maintained for both the His- principles. Each one of these states, in turn, has tory Tool and the Orbit Window. If the student separate subgoals which may or may not be met. The demonstrates inefficient behavior as indicated by one of tutor uses the State Module in two ways. First, if the the effectiveness measures, then the Coach intervenes student is performing poorly, then the Coach checks to and offers advice. see if the student has proceeded through each state in an appropriate manner. Second, the Coach uses the G. The Coach State Module to reflect different “instructional” stra- The Coach maintains the rules and procedures tegies. For example, if the student is conducting experi- that direct the teaching portion of the Tutor. The ments (as defined as “making predictions”) then the Ground Track Microworld is designed for two major system gives a higher status to using tools correctly. If purposes: 1) to teach students about the relationships the student is “testing,” then the Coach will switch its between/among orbital elements and ground tracks, strategy and try rules that check for skill deficiencies. and 2) to teach students how to use systematic behaviors to investigate this domain Thus, the Coach intervenes when either one of these conditions is not The major purpose of the Diagnostician is to analyze satisfied. The Coach monitors the student’s actions and student’s responses and update the Student and Curri- determines when the student needs advice. Interven- culum Models. Whenever the student enters a predic- tion occurs only when the student is making erroneous tion from the Prediction Window or changes parame- predictions or entering incorrect parameters in the Test ters from the Testing environment, the Diagnostician Mode. The general or high-level teaching strategy for compares the student’s answer to the Expert’s answer the Coach is as follows: and determines exactly which rules the student under- If the student has made No errors stands and does not understand. This information is and if the student is completing curriculum then transmitted to the Student Module which, in materials efficiently turn, stores it for further processing. then record progress The Diagnostician is also responsible for identify- ing the student’s errors and ill-defined strategies. The If the student has made No errors Diagnostician does this by combining information and if the student is NOT completing the obtained from the Expert Module, History files, and a curriculum materials efficiently series of high-level rules that generate students’ errors. then recommend an easier curriculum For example, if the student enters an erroneous predic- _tion for the orientation shape-descriptor, the Diagnosti-

Swigger, Burns, Loveland, and jackson 75 ‘87 with students from the Space School at Lowry Air If student has made error Force Base and from the Air Force Academy to deter- then mine if the tutor is more effective than traditional class- a) Check ruleset for satisfaction of room experience. This data will also be used to improve preconditions the diagnostic portion of the tutor. b Check ruleset for Correct Tool Use Several areas of research are also being investi- c I Check ruleset for Skill remediations gated using the ground track domain. The intelligent The authors made the general assumption that when tutor for this domain closely resembles an intelligent the student is in the Prediction Mode, then the Coach tutoring system developed by Schute and Glaser [1987] should help students discover the objectives by having which is currently used at Lackland Air Force Base, them use the tools correctly. If thii fails, then the sys- San Antonio, Texas, to identify individual cognitive tem should address individual skill errors. Thii strategy differences among students. We are planning to test is reversed whenever the student enters the Testing the effectiveness of the acquisition of inquiry skills by State. comparing Airmen who use both the Shute & Glaser The Coach’s overall intervention strategy is to Economics Tutor and the Ground Track Tutor. From check whether the student has completed the necessary this data, we will be able to determine the extent to preconditions (as determined by the values stored in which individuals transfer experimental behavior. the State Module). If the student has satisfied all the Because one of the primary purposes of this tutor was preconditions for an exercise, then the Coach checks to create a vehicle for testing hypotheses for training the measures for effective inquiry skills. The lit of effec- effectiveness, we want to investigate specific questions tive inquiry skills as originally defined in Shute and dealing with this area such as: What happens in an Glaser [19871 include: Systematic experimental instructional environment when you vary the order of behaviors such as making sufficiently large/small incre- the State Module? (Is it better to state a hypothesis ments to orbital parameters; Inductive/generalization and then conduct experiments?) What happens in the strategies such as replacating a test or prediction; Com- instructional environment when you vary the order of plexity of data organization such as isolating similar remediation? (Tool use versus Skill Diagnosis?) Finally, traces in the History file, selecting relevant ground how can the information we obtain from these studies traces in the History file; Strategies for diiconfirming be made a dynamic part of the system so that it can evidence such as re-doing the experiment or adjusting adapt to individual student’s needs? These and other orbital parameters to fit a new prediction. issues will be explored in the coming months and Every time a student enters a prediction or esti- should contribute to our understanding of how to build mates the orbital parameters in the Test Mode, the more effective training systems. Coach evaluates the Student Model and determines if intervention is required. If the student’s effectiveness measures are low, then the Coach proposes possible Acknowledgements remediation and offers assistance. In the event that the The authors gratefully acknowledge Dr. Valerie Schute student fails to attain a level of proficiency after receiv- whose contributions to this project were invaluable. ing instruction on effective Tool Use, then the Coach The author would also like to thank the instructors and addresses the student’s domain knowledge inadequa- students at the Unified Space Training School (UST) cies. for their assistance with this project. At the present time, the Coach uses the informa- tion stored in both the Tool Objects and the Expert References Module to advise the student concerning errors. Ini- [Astronautics, 19851 Astronautics 332. USAF Academy, tially, the system suggests that the student use one of Colorado, 1985. the available tools to correct hii errors. If the student continues to have difficulty, then the Coach may [Bate et al., 19711 Roger R. Bate, Donad D. Mueller, display the definitions, examples or explicitly state the and Jerry E. White. Fundamentals of Astrodynamics. relationships between various parameters. Dover Publications, New York, 1971. (Carr and Goldstein, 19771 B.Carr and Ira Goldstein. III. Summ~ and Puture Overlays: a theory of modelling for Computer Aided Instruction. MIT AI Memo 406 Memo 40. The current ground track microworld uses a qualitia- tive model to teach the basic concepts of orbital [Shute and Glaser, 19871 Valerie Shute and Robert mechanics. Thii microworld provides the student with Glazer. An Intelligent Tutoring System for Exploring a discovery environment which allows hi to explore Principles of Economics. In Press. relationships between orbital parameters and ground tracks. The microworld also has intelligence. It knows about the domain, about how to estimate orbital parameters from a ground track, and about how to use the inquiry tools effectively to achieve goals. As a result, if the student fails to make satisfactory progress toward the stated goals, then the system intervenes and offers appropriate assistance. Thii type of intelli- gent simulation provides a more active and adaptive environment for reinforcing training skills. The initial prototype is now complete and has been formatively evaluated by members of the NORAD crew and instructors at the Space School. The authors performed further tests during the Spring Semester of

76 Al & Education