Experimental Test report
Document information Project Title 6th Sense Project Number E.02.25 Project Manager Fraunhofer Austria Deliverable Name Verification Report “Experimental Test Report” Deliverable ID Del 4.2 Edition 00.01.01 Template Version 03.00.00 Task contributors
Fraunhofer Austria, FREQUENTIS AG, Fraunhofer FKIE, subcontracted by Fraunhofer Austria
Abstract The project Sixth Sense postulates that the users “body language" is different at “good” and “bad” decisions. The project follows the idea to use the whole body language of a user for communicating with a machine. In our case it is an Air Traffic Controller (ATCO) with an Air Traffic Tower CWP. Specifically we intend to analyse the correlation of the change of the behaviour of an ATCO - expressed through his body language - with the quality of the decisions he is making. For that, an experiment was set up, data about the user behaviour was collected, explored and analysed. This document is the test report of the proof of concept for the Sixth Sense prototype and its core components. Results of our work may be used for an early warning for “bad” situations about to occur or decision aids for the ATCO. Authoring & Approval
Prepared By - Authors of the document. Name & Company Position & Title Date Volker Settgast/Fraunhofer Austria Project Contributor 22.06.2015 Nelson Silva / Fraunhofer Austria Project Contributor 01.07.2015 Carsten Winkelholz, Jesscia Schwarz/ Subcontracted Project 07.07.2015 Fraunhofer FKIE Contributor Michael Poiger / Frequentis AG Project Contributor 08.07.2015 Florian Grill / Frequentis AG Project Contributor 08.07.2015
Reviewed By - Reviewers internal to the project. Name & Company Position & Title Date Theodor Zeh / Frequentis Technical Coordinator 15.07.2015
Eva Eggeling / Fraunhofer Austria Project Manager 15.07.2015
Approved for submission to the SJU By - Representatives of the company involved in the project. Name & Company Position & Title Date Theodor Zeh / Frequentis Technical Coordinator 30.07.2015
Eva Eggeling / Fraunhofer Austria Project Manager 30.07.20125
Rational for rejection None.
Document History
Edition Date Status Author Justification 00.00.01 20/06/2015 Draft Eva Eggeling New Document 00.00.03 07/07/2015 Update Volker Settgast merged version 00.00.04 09/07/2015 Update all merged version 00.00.08 15/07/2015 Update all merged version Submission 00.01.00 30/07/2015 Eva Eggeling Merged Version Version 00.01.01 15/09/2015 Review Version Eva Eggeling/all Resubmission
Intellectual Property Rights (foreground) This deliverable consists of SJU foreground.
2 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Table of Contents TABLE OF CONTENTS ...... 3 LIST OF TABLES ...... 4 LIST OF FIGURES ...... 4 EXECUTIVE SUMMARY ...... 6 1.1 PURPOSE OF THE DOCUMENT ...... 7 1.2 INTENDED READERSHIP...... 7 1.3 ACRONYMS AND TERMINOLOGY ...... 7 2 THE EXPERIMENT ...... 9 2.1 EXPERIMENTAL SETUP ...... 9 2.2 OPERATIONAL SCENARIO ...... 10 2.3 ROLES AND RESPONSIBILITIES ...... 11 2.4 TECHNICAL SETUP OF THE EXPERIMENT ...... 12 2.5 AMQ BROKER ...... 13 2.6 THE HUMAN MACHINE INTERFACE (HMI) ...... 13 3 PERFORMING THE EXERCISES ...... 16 3.1 PROFILE OF PARTICIPANTS ...... 16 3.2 DATA ANALYSIS, EXPLORATION AND VISUALIZATION ...... 17 3.2.1 Heart Rate vs Observations List ...... 18 3.2.2 Eye-Tracker and mouse Analysis ...... 19 3.2.3 Simple Metrics and Data Exploration ...... 20 4 RESULTS ...... 23 4.1 WORKLOAD ESTIMATES BASED ON QUESTIONNAIRES ...... 23 4.2 HINTS FOR H1 - EXPLORING THE SENSOR DATA ...... 28 4.2.1 Sixth Sense Prototype Framework for Data Exploration ...... 31 4.2.2 Categorization of Metrics regarding mental Aspects ...... 34 4.2.3 Research Questions ...... 38 4.2.4 Data Exploration and Analysis ...... 40 4.3 HINTS FOR H2 - ANALYSIS OF THE ARRIVAL AND DEPARTURE WORKFLOWS ...... 54 4.3.1 Implementation ...... 54 4.3.2 Results of the Analysis of ATC Workflow Steps ...... 55 4.3.3 Machine Learning Experiments ...... 57 4.4 EVENT TRACE ANALYSIS ...... 61 4.4.1 Variable Length Markov Models (VLMM) ...... 61 4.4.2 Scatterplot Matrix for Measures ...... 64 4.4.3 Visualization of Sequential Patterns ...... 64 4.4.4 Insights regarding interaction sequences ...... 66 4.4.5 States corresponding to outliers and around ...... 75 4.5 CONCLUSION ...... 77 4.6 FUTURE WORK ...... 79 REFERENCES ...... 80 APPENDIX A ...... 82 A.1 TECHNICAL VERIFICATION DETAILS OF EXERCISE 1 AND 2 ...... 82 A.1.1 Kinect ...... 87 A.1.2 Speech Recognition ...... 92 APPENDIX B QUESTIONNAIRES ...... 94
3 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
List of tables Table 1 - Description of the workflow steps ...... 9 Table 2- Data collection and Quality Assessment for Different Data Sets and Sensors ...... 17 Table 3 - Resume of initial metrics to be visualized and explored ...... 21 Table 4 - Outliers in negative/positive answers...... 28 Table 5 - Resume of most important metrics ...... 31 Table 6 - Classification of most important metrics into categories...... 35 Table 7- List of Main Research Questions ...... 39 Table 8 - Resume of AOI that received most interest time from each user...... 45 Table 9 - Resume of parameters for the Kinect Head Pose ...... 48 Table 10 – Filter/Query to detect airplanes that are in the workflow step TAXI...... 55 Table 11 - Most frequent state sequences for the eye data (top5 for each user)...... 67 Table 12 - Most frequent states of each user for the eye fixation sequences...... 68 Table 13 - Most frequent state sequences for the mouse data (top 5 for each user)...... 69 Table 14 - Most complex state sequences for the eye tracking data (top 5 for each user) ...... 70 Table 15 - Illustration of the most complex state sequences for the eye tracking data ...... 72 Table 16 - Most complex state sequences for the mouse data (top 5 for each user)...... 73 Table 17 - Illustration of the most complex state sequences for the mouse data ...... 74 Table 18 - Examples of state sequences corresponding to outliers in the scatterplots...... 76 Table 19 -Technical specifications of Kinect...... 88 Table 20 - Kinect Results ...... 91
List of figures Figure 1 - Update of the exercise plan as described in the experimental plan...... 9 Figure 2 - Experimental Workflow ...... 10 Figure 3 - Hamburg Airport ...... 11 Figure 4 - Arrival workflow - responsibilities ...... 12 Figure 5 - Departure workflow - responsibilities ...... 12 Figure 6 - Setup working position ...... 13 Figure 7 - Components of the HMI screen ...... 14 Figure 8 - Departure Strips ...... 14 Figure 9 - Arrival Strips ...... 15 Figure 10 - Strip Bay Configuration / Button Bar...... 15 Figure 11 - RMSSD Formula...... 18 Figure 12 - Z-Score IBI vs negative observations through the total experiment time for user8...... 19 Figure 13 - Areas of interest of the ATC Simulator as defined in Ogama ...... 20 Figure 14 - Ranking of metrics and visualizations ...... 21 Figure 15 - Observation List ...... 22 Figure 16 - Mental Demand Results for all 8 users, 2 experiments...... 24 Figure 17 - Physical Demand Results for all 8 users, 2 experiments...... 24 Figure 18 - Temporal Demand Results for all 8 users, 2 experiments...... 24 Figure 19 - Level of Effort Results for all 8 users, 2 experiments...... 24 Figure 20 - Level of Frustration Results for all 8 users, 2 experiments...... 25 Figure 21 - NASA-TLX Negative Results (not considering “Level of Performance” answers)...... 25 Figure 22 - Level of Performance (for all users, 2 experiments)...... 25 Figure 23 - NASA-TLX Correlation Matrix (taking all answers from all users)...... 25 Figure 24 - SAGAT Based Questionnaire...... 26 Figure 25 - SAGAT Correlated Answers...... 26 Figure 26 - SASHA based Questionnaire...... 27 Figure 27 - SASHA Questionnaires, correlated Plot...... 27 Figure 28 - NASA-TLX and SASHA Correlation Matrix...... 27 Figure 29 - Negative vs Positive Answers (based on all questionnaires)...... 27 Figure 30 - Overview of the Sixth Sense Desktop Application Prototype ...... 31 Figure 31 - Screenshot of the Sixth Sense desktop application UIAction Pace Calculator...... 33 Figure 32 - Sixth Sense desktop application UI Actions Types Monitor...... 33
4 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Figure 33 - Sixth Sense web based reports for supervisors data exploration, also printable...... 34 Figure 34 - Distinction between task load and workload (Hilburn & Jorna, 2001) ...... 35 Figure 35 - Relationship between workload and performance (Veltman & Jansen, 2003) ...... 36 Figure 36 - Events from observation list with high impact on the performance of the users...... 37 Figure 37 - Events from observation list with more impact on the performance of each user...... 38 Figure 38 - Departures and arrivals (green area) vs number of negative observations (red) ...... 40 Figure 39 - Interdependence between arriving airplanes, departures and stress levels...... 41 Figure 40 - Correlation between negative observations and HRV. HRV is a good indicator for periods of negative observations...... 42 Figure 41 - Relation between mouse AOI frequencies observation list and HRV...... 43 Figure 42 - Mouse AOI of user7 that received most interest time during the experiment...... 44 Figure 43 - Eye AOI of user7 that received most interest time during the experiment...... 44 Figure 44 - Window standard deviation o(2 minutes) and number of errors, capturing very well periods with increased user errors...... 46 Figure 45 - Relation between eye and mouse movements (AOI visits) and occurrence of errors...... 47 Figure 46 - Kinect Head Pose Measurements Schema...... 48 Figure 47 - Kinect Data Representation, visualizing Detected Head Pose vs Count of Negative/Positive Observations vs Type of Observation vs User in Range (or not in range)...... 49 Figure 48 - Kinect Data after applying filters to include only the majority of negative observations (96%)...... 50 Figure 49 - Correlation between total number of mouse clicks and negative observations ...... 51 Figure 50 - Correlation of negative observations and difference in number of words/mouse actions .. 52 Figure 51 - Relationship between number of words spoken and negative observations...... 53 Figure 52 - Example of using CEP to join two different events into one...... 54 Figure 53 - The complete process of consuming, filtering and generating events...... 55 Figure 54 - Analysis of the Processing Time (seconds) for arrivals (orange/brown) and Departures (Blue) for user8...... 56 Figure 55 - DM/ML/AI module with automatically calculated metrics for arrival flights capturing repeated workflow steps (e.g., number of taxi commands or cross runways for all flight)...... 57 Figure 56 - Discovery of Association Rules using the algorithm fp-growth...... 58 Figure 57 - Relation Between the discovered association rules and different variables of the model. 58 Figure 58 - Outliers Discovery for negative observations in the new dataset with metrics counters (captured between successive negative observations)...... 59 Figure 59 - Decision tree to depict reasons for increasing numbers of negative occurrences for different users...... 60 Figure 60 - Polynomial regression analysis for creating a model to predict negative occurrences based on top most metrics (number of eye events or departure flights)...... 61 Figure 61 - Relation of Probabilistic Suffix Tree (PST) and Automation PSA...... 62 Figure 62 - Hypothetical distribution of event durations, if after one event (left) or a sequence of two events (right) a specific event is observed...... 62 Figure 63 - Illustration of the complexity measure of Grassberger...... 63 Figure 64 - Screenshot of the user interface with displayed transition probabilities...... 65 Figure 65 - Illustration of how probabilities for next events in a sequence are displayed...... 65 Figure 66 - Illustration of the user interface combining states with displayed scatterplot matrix...... 66 Figure 67 - Eye-Tracking ...... 82 Figure 68 - Test Setup - Eye-Tracking ...... 83 Figure 69 - Eye Tracking Data Analysis ...... 84 Figure 70 - Test Person 1 – Eye Tracking ...... 85 Figure 71 - Test Person 2 – Eye Tracking ...... 86 Figure 72 - Test Person 3 – Eye Tracking ...... 86 Figure 73 - Test Person 4 – Eye Tracking ...... 87 Figure 74 - Kinect sensor ...... 87 Figure 75 - Sensors included in the Kinect ...... 88 Figure 76 - Test Setup - Kinect ...... 89 Figure 77 - Evaluation of distances and angle ...... 90 Figure 78 - Test Setup – Speech Recognition ...... 92 Figure 79 - Callsign Recognition Rate – Speech Recognition ...... 93
5 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Executive summary The project Sixth Sense follows the idea of using the whole body language of a user for communicating with a machine. In our case it is an Air Traffic Controller (ATCO) with an Air Traffic Tower CWP. Specifically we intend to analyse the correlation of the change of the behaviour of an ATCO - expressed through her/his body language - with the quality of the decisions she/he is making. Result of our work may be used for an early warning for “bad” situations about to occur or decision aids for the ATCO. We used scenarios of Hamburg Airport since its layout has sufficient complexity to bring the test personnel in difficult situations which are needed to test our hypothesis. Sensors for reading the body language were: Kinect for body movement Eye tracking for gaze detection Speech recognition Mouse cursor position Room temperature Heartbeat of user Expert observations The sensors were recorded through each run together with the workflow/the tasks performed by the user. The workflow was retrospectively analysed by experts who marked bad decisions and/or bad situations arising. Combinations of sensor recordings and different visualisations therefrom were used to detect repetitive patterns of user behaviour correlating to good or bad decisions. Several test runs were performed in two batches to gain as much test data as possible in the available time frame to experiment with. Details are in the paragraphs below.
Key learnings of the work performed were: An analysis of decision quality through experts is difficult since the intention of the test person stays hidden. Additional self-assessment will add value in future tests. Analyses of sensor recordings offer infinite possibilities of combinations as well as visualisations therefrom. Further work on the existing data might produce even more significant findings
Conclusion: our test setup and process proofed right. Analytical tools and visualisations used are feasible although there are numerous other possibilities which might be even better. Due to the nature of this kind of exploratory research projects with restricted resources no statistical relevance in the found patterns is recognisable. The number of test persons was too low. However, the concrete patterns which have been found allow deriving early indications for good or bad decisions. There are good indications for positive results when more test data and more time is available for sensor permutation analysis.
6 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
1.1 Purpose of the document This document provides a results report to the experiment in the Sixth Sense project. Chapter 2 describes the setup of the experiment and explains the exercises. The performance of the experiment is summed up in Chapter 3. In Chapter 4 we present the results: We give a classification of our most important metrics related with task-load, mental workload, attention, behaviour and performance. Then we have a deeper look into related research questions and describe the complex data analysis. Relationships between the sensor data streams are discussed and conclusions of the analysis are described in detail. We explain the capabilities of our software framework and explain future directions of research, for example the use of the current results to create predictive models or how to make improvements in the user interface to support the user in making more informed decisions. 1.2 Intended readership This document might be of interest for: Sixth Sense project members, including the project manager and the core team members. Representatives of EUROCONTROL and SJU responsible for reviewing and advising the project. Other researchers working on the related research projects, particularly researchers on error avoidance, new technologies and interaction methods. Personnel in air traffic management and other parts of the aviation sector. 1.3 Acronyms and Terminology
Term Definition
AMQ Active Message Queue
ARR Arrival
ATCO Air Traffic Control Officer
ATM Air Traffic Management
DEP Departure
DM Data Mining
IBI Inter Beat Interval
MFA Multilateral Framework Agreements
HRV heart rate variability
KPI Key Performance Indicator
ML Machine Learning
Negative error Negative situation that could not be resolved
7 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Term Definition
Positive error Negative situation, which could be solved with effort of the user
SESAR Single European Sky ATM Research Programme
SESAR Programme The programme which defines the Research and Development activities and Projects for the SJU.
SJU SESAR Joint Undertaking (Agency of the European Commission)
SJU Work Programme The programme which addresses all activities of the SESAR Joint Undertaking Agency.
TWR Tower
.
8 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
2 The Experiment This section provides general information on the final design of the experiment, the preparation of the exercises and their performance. In contrast to the original plan described in the experimental plan 4.1, due to limited resources we skipped the AI-module development and did not perform Exercise 3. A few tasks of Exercise 3 (first steps towards prediction) were handled by analysing the data collected in Exercise 2.
Excercise 1 Excercise 2 Excercise 3
Expert KPIs ratings
Test of Development of Predictions of Expert KPIs Tracking Sensors DM/ML/AI‐module DM/ML/AI‐module ratings
Test of DM/ML/AI‐module
Figure 1 - Update of the exercise plan as described in the experimental plan. The prediction and DM/ML/AI-module-test, was part of Exercise 3 and could not be processed because of the limited amount of time, resources and data. First steps regarding predictions are described in Section 4.3.3. 2.1 Experimental Setup The following steps have been conducted to execute the experiment where a participant performs a simulated 60 minutes ground controller shift at a simulated ground controller position.
Name Description
Overall Experimental Provision of an overall briefing, providing an overview about the used Briefing system and the operational scenario conducted during the exercise.
Start of Experiment Reset of the operational scenario.
A_Pre-Questionnaire Collecting information about test person (working experience, etc.).
Recording of data Start of the recording of data, to be collected into the database.
Run Exercise Start of the operational scenario and conducting the exercise
B_Supervisor Observation During the exercise the observer took notes and collected the stress level
C_Post-Questionnaire Collection about subjective feelings (situational awareness, workload).
D_Debriefing Collection of debriefing questionnaire answers of the test person.
Overall Experiment General Debriefing to close the experiment session. Debriefing Table 1 - Description of the workflow steps 9 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Every participant received a map of the airport (Hamburg) and was asked to assume the experiment work place. The participants were informed that they could ask questions about the use of the simulator user interface to the air traffic controller supervisor, present in the room. When all questions were answered, the experiment started, and the air traffic information was loaded into the simulator. Every 10 minutes, the supervisor asked, what was the current stress level experienced by the participant and registered extra notes about his personal evaluation point of view of the current performance of the participant. The experiment lasted for 45 minutes, but it could run for 60 minutes maximum, depending on the current air traffic situation.
Overall Experiment Briefing (incl. Operational)
Start of Experiment
A_Pre‐Questionnaire
B_Supervisor Observation Recording of data Run Exercise (Observation List)
C_Post‐Questionnaire
D_Debriefing
Overall Experiment Debriefing
Figure 2 - Experimental Workflow
Table 1 and Figure 2 provide an overview of the different workflow steps within the experimental scenario. The Questionnaires A-D can be found in Appendix B. 2.2 Operational Scenario The operational scenario was based on Hamburg Airport.
10 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Figure 3 - Hamburg Airport Following constraints have been used to prepare the scenario: Simulation prepared for approx. 60 min. Arrivals are automatically simulated until touchdown. (no change of route) Departures are controlled until take off. No Runway change is foreseen within the simulation Taxiway Routes can be selected by the operator.
Configurations during the experiment: Arrival Runway: 23 Departure Runway: 33 Arrivals: 31 flights Departures: 27 flights
2.3 Roles and Responsibilities The following roles were participating in the experiment: Ground Controller: Participant Runway Controller: Manually Simulated Pseudo Pilots: Manually Simulated Observer (supervisor) Observer (experiment leader)
11 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
The responsibilities within the workflow are displayed in the following Figures:
Figure 4 - Arrival workflow - responsibilities
Figure 5 - Departure workflow - responsibilities
2.4 Technical Setup of the experiment The setup is based on a single simulated controller working position. No 3D view is available at the experiment. The experiment concentrated on the ground traffic management.
The following modules have been used during the experiment: Traffic Simulator CWP with EFS, Support Information AMQ Broker Eye-Tracker Mouse Keyboard Speech Recognition
12 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Experimental CWP
Simulator Position Ground Position Observer Position
Voice Recognition Ground Controller Position Video / Screen Capturing Server Components AMQ Broker Simulator MySQL Data Logger RWY Controller Position Pilot Simulation
Figure 6 - Setup working position 2.5 AMQ Broker The broker is the central distribution system for all data communication between all components. The transport protocol used is stomp and open wire. ActiveMQ (AMQ) – allows a single point of data exchange between different systems, modules and functional blocks, through the usage of customized XML messages. It supports a variety of cross language clients and prototcols from Java, C, C++, C#, Python … For detailed information please refer to: http://activemq.apache.org/ 2.6 The Human Machine Interface (HMI) The HMI is split into several parts which are described in more detail in the following paragraphs. In general, the right side of the screen is reserved for a representation of the flights, the smartStrips. The middle part can contain a variety of different information including an overview of the airfield as shown in the picture below. The top contains an information bar and to the side there is a sidebar containing additional information e.g. status of the system or wind data. Summarized, the screen can thus be separated into: Sidebar Infobar Button bar (EFS) Strips Page Selection (Main Area)
13 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Figure 7 - Components of the HMI screen Figure 8 explains the different fields of the departure strip. Three different sizes are available: DEP MICRO, DEP MEDIUM, and DEP MACRO. The filled explanations mean that this field can be pressed on the strip. Figure 9 explains the fields of the arrival strip. Three sizes are available: ARR MICRO, ARR MEDIUM, and ARR MACRO. The filled explanations mean that this field can be pressed on the strip.
Figure 8 - Departure Strips
14 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Figure 9 - Arrival Strips
Figure 10 - Strip Bay Configuration / Button Bar In Figure 10 the information about the configured bays and the explanation of the button bar are shown.
15 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
3 Performing the Exercises The first exercise of this experiment was used to assess the accuracy of the sensors integrated into the prototype to ensure the necessary quality of the used technologies. The following sensors were tested in Exercise 1: ‐ Eye Tracker ‐ Kinect ‐ Speech Recognition See Appendix A, Exercise 1 of the deliverable D4.2 (Verification Plan): Exercise ID/Title: EXE-E.02.25-VP-0001.0001 /Eye-Tracking adapter Exercise ID/Title: EXE-E.02.25-VP-0001.0002 /Kinect AMQ-adapter Exercise ID/Title: EXE-E.02.25-VP-0001.0003 /Leap Motion AMQ-adapter The Leap Motion was found to be not useful in a sitting mouse environment. Instead, the speech recognition module was evaluated in more detail. More technical details about exercise 1 can be found in Appendix A of this document.
In Exercise 2 the participants followed the experimental workflow of Figure 2 and performed the simulated 60 minutes shift of a ground controller. During the exercise, the supervisor took observation notes and asked the participant for her/his stress level (on a scale from 1 to 5) every ten minutes. The observation notes consist of a time stamp and a short description of the observation. In the second part of Exercise 2 (user 5-8) this process was already automated and stored using the software framework. In a later step, the recorded video capture of the exercise was revised by a domain expert to create the observer list. The observer list consists of selected events which are rated positive, neutral and negative. A positive event occurs when the participant can successfully resolve a negative event. The detailed description of exercise 2 can be found in Appendix B Exercise 2 of the deliverable D4.2 (Verification Plan): Exercise ID/Title: EXE-E.02.25-VP-0002.0001 / Collecting Sensor data, and expert reviews. The questionnaires before, during and after the exercise can be found in Appendix. As mentioned in Section 2, in contrast to the original plan, we did not perform Exercise 3, we could only do the first steps regarding predictions in the data analysis of Exercise 2 data. 3.1 Profile of Participants All participants work in the field of air traffic control but at different expert levels: as air traffic controllers, one En-route, two Ground, one trained as a ground controller but works only in simulations experiments. Years of work experience: The years of professional experience differ between 2, 4, 14 and 20 years respectively. Gender: There were two male (50%) and two female (50%) participants. Age: One participant was aged between 20 and 30, one was aged between 30 and 40 and two were aged between 40 and 50. Language: Two of the participants had German as their mother language, one Romanian and the other Spanish. All communications between pilots and air traffic controllers were handled in English.
16 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged.
Due to limited resources of test participants we had to reuse participants for the experiments. Learning effects caused by this reuse may not be ruled out but as we tried to measure behaviour for individual test runs, this effect can be neglected. 3.2 Data Analysis, Exploration and Visualization After performing Exercise 2 and post-processing of the data, a resume table (see Table 2) with the total number of usable events for each generated dataset (topic) was created.
Data Assessment
Total Topics User User User User User User User User Events Descriptio Variables (Datasets) 1 2 3 4 5 6 7 8 by n Topic Reports from Supervisor & Supervisors 3 51 65 13 123 91 91 107 152 693 Observer and Observers Stress Level 3 StressLevel 6 6 6 6 6 6 7 6 49 reports (from users) Flight 42 FlighObject 616 420 57 436 302 241 340 434 2846 Information Strips 6 Selections 420 302 17 211 197 149 209 268 1773 Selections 10 Eye 0 0 53 169 9097 61770 72534 68116 211739 Eye Tracker Mouse UI 4 GlobalMouse 0 72891 2929 79618 3844 12700 7588 8082 187652 Hook Mouse 7 Mouse 3046 1929 110 1290 916 1838 1266 1915 12310 Listner Kinect 23 Kinect 27351 0 0 0 7561 0 0 0 34912 Listner 12 Voice 1014 1160 36 1242 256 899 1126 1587 7320 Voice Listner Waspmote 4 Waspmote 0 0 0 0 1807 2754 3041 3376 10978 Listner Heart Rate Heart Rate 12 Measurement 0 2978 2274 5347 0 4512 0 5184 20295 Events s Collected Eye AOI Eye Tracking (Fixations, 13 0 0 0 0 715 1625 3396 3767 8788 Areas of Gazes, Interest Sacades) Mouse AOI Mouse (Fixations, Tracking 13 0 2217 0 2325 62 2729 1002 1083 9418 Gazes, Areas of Sacades) Interest Total Data <= Total Collected in Number of 32504 81968 5495 90767 24139 89314 90616 93970 152 508773 2 Variables experiments Number <= Total of Total Number of Number of Events Events 58 Airplanes Collecte d
27 <= Departures 31 <= Arrivals
Table 2- Data collection and Quality Assessment for Different Data Sets and Sensors
The first entry in Table 2 we entered the notes and annotations from the supervisor and observer of the exercise (see also Section 3.2.3 “Observations”). These notes are text notes about for example observations of errors or suboptimal situations. When talking about observations of errors we
17 of 98
©SESAR JOINT UNDERTAKING, 2011. Created by Fraunhofer Austria, FREQUENTIS AG for the SESAR Joint Undertaking within the frame of the SESAR Programme co-financed by the EU and EUROCONTROL. Reprint with approval of publisher and the source properly acknowledged. distinguish between positive and negative, where positive means that negative situations, could be resolved by some effort of the user.
The second entry is the stress level. We acquired this information by asking the participant every ten minutes about the subjective stress level on a scale from 1 to 5.
During the exercises we encountered several hardware issues with the Kinect sensor. In favour of a higher eye tracking frequency it was decided to deactivate the Kinect for user 6-8.
3.2.1 Heart Rate vs Observations List Not all the users agreed in wearing the heart rate monitor device (for different reasons health, privacy). For user2, user3, user4, user6 and user8 we collected at least 3 baseline measurements at rest. In addition to the heart beat per minute we measured also the heart rate variability (HRV). The HRV indicates the fluctuations of the heart rate around an average heart rate. An average heart rate of 60 beats per minute (bpm) does not mean that the interval between successive heartbeats would be exactly 1.0 sec. Instead the interval may fluctuate/vary from 0.5 sec up to 2.0 sec. HRV is affected by aerobic fitness and HRV of a well-conditioned heart is generally large at rest. Other factors that affect HRV are age, genetics, body position, time of day, and health status. During exercise, HRV decreases as heart rate and exercise intensity increase. HRV also decreases during periods of mental stress. The HRV is regulated by the autonomic nervous system. Parasympathetic activity decreases heart rate and increases HRV, whereas sympathetic activity increases heart rate and decreases variability. A low HRV indicates dominance of the sympathetic response, the fight or flight side of the nervous system associated with stress, overtraining, and inflammation. Therein lies the beauty of HRV: it offers a glimpse into the activity of our autonomic nervous system, an aspect of our physiology normally shrouded in mystery. For the representation of the HRV we may use the time-domain method of the Root Mean Square of the Successive Differences (or RMSSD) as we can see in Figure 11: HR = heart rate in beat per minute (bpm) ) = no. of R’s R - R interval = inter-beat interval (IBI) in msec. N = no. of R - R interval terms