PRE2017 3 Groep16

From Control Systems Technology Group

Jump to: navigation, search

Contents

Group Members

  • Rivelino Wattimena - 0967390
  • Jeroen van Meurs - 0946114
  • Tim Driessen - 0954562
  • Jorik Mols - 0851883
  • Lisanne Willems - 0954451

Final Deliverable

We made a report style deliverable. This can be found via the following link: Final Report

Contents

This wiki serves as a logbook to show progress that is gained, and changes that have been made each week. A final research report can be found externally.

Week 1

Problem statement

Imagine a world in which autonomous vehicles fill the streets. For a lot of people this would be ideal traffic control. However, no human control over the car brings some setbacks. One of these setbacks is interaction with pedestrians that are crossing the road. In today's traffic, a car driver will usually wave at the pedestrians, to show that they have been seen and they can cross the road. However, with autonomous self driving cars, no humans are in control of the vehicle. Then how do pedestrians know that the vehicle has seen them? Also the opposite case is important, how do autonomous vehicles know the pedestrians have seen them? This human-vehicle interaction problem is of great importance for the general safety in trafic. For this problem, we will try to come up with some solutions.

Objectives: Create a safer environment An interaction between autonomous vehicles and pedestrians that increases safety (for all involved) in traffic.

Users

The system will have a set of users, each with their own requirements. These might vary among the different user types. The pedestrians are a group of users that will indirectly make use of the system, by interacting with the autonomous vehicle which has incorporated this system. These pedestrians are concerned about safety and will want to trust that this interaction does not fail. When wanting to cross the road, they should not have to perform actions that are too complex, so the system should be easy to work with (ease-of-use). The driver of the autonomous vehicle (or rather, passenger) also wants to be able to trust this system as well as the autonomous vehicle itself. Since we target fully autonomous vehicles and not vehicles that still require some control of the driver, we envision the passengers of such a vehicle to trust what the vehicle is doing. The driver is also concerned with safety, accidents are to be avoided of course. Our system should be able to deal with all necessary interaction between the vehicle and the pedestrian, therefore the driver might not have to be involved in this interaction. We will have to determine whether this is the case when designing our system. The autonomous vehicle itself also counts as a user (even though it is not human). The workings of these vehicles should be improved with our solution and in traffic (autonomous) vehicle-pedestrian interaction should be safer.

Society

Our society should benefit from our solution. Governments spend millions of dollars already to increase traffic safety. Although we are definitely not at the stage where everyone drives an autonomous vehicle, we expect this to be the future and safety is always a concern when it comes to traffic. What we have to research is where exactly our system will be a solution to the problem. Pedestrian-dense neighborhoods where people are used to crossing roads with little care might require a different tactic then places where there are a lot of pedestrian crossings which are properly used. Autonomous vehicles should be risk-averse and thus might be too careful when driving in environments like the center of Amsterdam, especially when we design a system that requires the vehicle to interact with each and every person that wants to cross the road.

Enterprise

Business-wise the system should be successful in that car production companies can buy and use it in their autonomous vehicles. To get autonomous vehicles more accepted by the public, they have to become safer so that people can trust them. Our system might have a positive effect on this, showing to the public that these autonomous vehicles can in fact be made safe. Car companies could market their cars with this idea and our solution in mind.

Approach

In this project we will determine the problems pedestrians face when crossing streets where autonomous vehicles drive and the other way around. Then we will look at numerous stakeholders and possible solutions. After that, questionnaires/interviews will be held with stakeholders to determine the needs of a system that offers a solution to the defined problem. After this, a design for our solution will be made and a prototype will show some of the working principles that need to be proven in order to give credibility to the final design.

Deliverables

We will write our findings in a report style on the wiki. We would like to deliver a prototype near the end of this project.

Who will do what?

For the first part of the project we will work together as much as possible so everybody has the same basis. In a later stage of the project we will split the work a little bit more. Since we have a student from software science, he will take the lead in the coding work. The mechanical engineers will take the lead in the hardware creation.

For this week the tasks are divided as the following:

  • Problem description: Tim Driessen
  • USE aspects: Jorik Mols, Jeroen van Meurs
  • State of the art research: Rivelino Wattimena, Lisanne Willems, Tim Driessen

State of the Art Research

  • Millard-Ball, A. (2018). Pedestrians, autonomous vehicles, and cities. Journal of Planning Education and Research, 38(1), 6-12. 10.1177/0739456X16675674

- Short Summary of Abstract:

In this article the author uses game theory to analyse interactions between pedestrians and autonomous vehicles with a focus on crossing streets.

  • Hulse, L. M., Xie, H., & Galea, E. R. (2018). Perceptions of autonomous vehicles: Relationships with road users, risk, gender and age. Safety Science, 102, 1-13. 10.1016/j.ssci.2017.10.001

- Short Summary of Abstract:

in this article a study is discussed where almost 1000 participants have been surveyed their perceptions, particularly regarding the safety and acceptance of autonomous vehicles.

  • Zhang, J., Vinkhuyzen, E., & Cefkin, M. (2018). Evaluation of an autonomous vehicle external communication system concept: A survey study10.1007/978-3-319-60441-1_63

- Short Summary of Abstract:

  • Chang, C. -., Toda, K., Sakamoto, D., & Igarashi, T. (2017). Eyes on a car: An interface design for communication between an autonomous car and a pedestrian. Paper presented at the AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Proceedings, 65-73. 10.1145/3122986.3122989

- Short Summary of Abstract:

In this article an interface design has been tested in VR for communication between autonomous cars and pedestrians. The evaluation results show that pedestrians can make the correct decision more quickly when the approaching car has the novel interface “eyes” than in case of a normal car. Furthermore the results also show that they feel safer crossing a street if the approaching car has eyes and if they make contact with the.

  • Mirnig, N., Perterer, N., Stollnberger, G., & Tscheligi, M. (2017). Three strategies for autonomous car-to-pedestrian communication: A survival guide. Paper presented at the ACM/IEEE International Conference on Human-Robot Interaction, 209-210. 10.1145/3029798.3038402

- Short Summary of Abstract:

in this article three strategies are discussed how autonomous cars could communicate with other agents for accident-free-traffic with the help of knowledge from social robots.

  • Rothenbücher, D., Li, J., Sirkin, D., Mok, B., & Ju, W. (2015). Ghost driver: A platform for investigating interactions between pedestrians and driverless vehicles. Paper presented at the Adjunct Proceedings of the 7th International Conference on Automotive User Interfaces and Interactive VehicularApplications, AutomotiveUI 2015, 44-49. 10.1145/2809730.2809755

- Short Summary of Abstract: In this article a simple test has been done to obtain how pedestrians will react to a “driverless vehicle”. A vehicle was prepared to make it look it was driverless and information could be obtained without really having an autonomous car.

  • Keferböck, F., & Riener, A. (2015). Strategies for negotiation between autonomous vehicles and pedestrians. Paper presented at the Mensch Und Computer 2015 - Workshop, 525-532. Retrieved from www.scopus.com

- Short Summary of Abstract:

In this article a study is discussed about comparing the actions of pedestrians with autonomous cars in two cases: when the car explicitly interacts with them or not explicitly interacts with them.

  • David, C., Wim, V., Ingrid, M., & Piet, D. (2011). Architecture for vulnerable road user collision prevention system (VRU-CPS), based on local communication. Paper presented at the 18th World Congress on Intelligent Transport Systems and ITS America Annual Meeting 2011, , 7 5500-5509. Retrieved from www.scopus.com

- Short Summary of Abstract:

In this article a proposition is made to use position estimation based on neighbouring devices such as other cars or smart devices.

  • Scaramuzza, D., Spinello, L., Triebel, R., & Siegwart, R. (2010). Key technologies for intelligent and safer cars - from motion estimation to predictive collision avoidance. Paper presented at the IEEE International Symposium on Industrial Electronics, 2803-2808. 10.1109/ISIE.2010.5636880

- Short Summary of Abstract:

In this article various techniques are discussed for safer autonomous driving in urban environments.

  • Colley, A., Häkkilä, J., Pfleging, B., & Alt, F. (2017). A design space for external displays on cars. Paper presented at the AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Adjunct Proceedings, 146-151. 10.1145/3131726.3131760

- Short Summary of Abstract:

In this article ideas are discussed to present information on the exterior of cars.

  • Schneemann, F., & Gohl, I. (2016). Analyzing driver-pedestrian interaction at crosswalks: A contribution to autonomous driving in urban environments. Paper presented at the IEEE Intelligent Vehicles Symposium, Proceedings, , 2016-August 38-43. 10.1109/IVS.2016.7535361

- Short Summary of Abstract:

In this article the interaction between drivers and pedestrians are analysed to define the behavioural requirements for future autonomous vehicles. A study has been conducted from both the driver’s perspective and the pedestrian’s perspective.

  • Saleh, K., Hossny, M., & Nahavandi, S. (2017). Towards trusted autonomous vehicles from vulnerable road users perspective. Paper presented at the 11th Annual IEEE International Systems Conference, SysCon 2017 - Proceedings, 10.1109/SYSCON.2017.7934782

- Short Summary of Abstract:

In this article a computation framework has been proposed for modelling trust between Vulnerable Road Users and autonomous vehicles based on a shared intent understanding between the two of them.

  • Wang, C. -., Liu, A., Wu, P., & Lu, P. -. (2017). A study in human-machine interaction through agent simulation: An application in pedestrian crossing. Paper presented at the 2016 International Automatic Control Conference, CACS 2016, 167-172. 10.1109/CACS.2016.7973903

- Short Summary of Abstract:

In this article research has been done by using agent simulation to realize Human-Vehicle interaction. The domain chosen is the Pedestrian-Vehicle in street crossing.

  • Gupta, S., Vasardani, M., & Winter, S. (2016). Conventionalized gestures for the interaction of people in traffic with autonomous vehicles. Paper presented at the Proceedings of the 9th ACM SIGSPATIAL International Workshop on Computational Transportation Science, IWCTS 2016, 55-60. 10.1145/3003965.3003967

- Short Summary of Abstract:

In this article the question is answered whether there is an universal language to interact with traffic.

  • Dey, D., & Terken, J. (2017). Pedestrian interaction with vehicles: Roles of explicit and implicit communication. Paper presented at the AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Proceedings, 109-113. 10.1145/3122986.3123009

- Short Summary of Abstract:

In this article road-crossing and communication behaviour of pedestrians and drivers in busy traffic situations are categorized. The evidence suggest that eye contact does not play a major role in manual driving and that motion patterns and behaviours of vehicles play a more significant role

  • Florentine, E., Andersen, H., Ang, M. A., Pendleton, S. D., Fu, G. M. J., & Ang, M. H. (2016). Self-driving vehicle acknowledgement of pedestrian presence conveyed via light-emitting diodes. Paper presented at the 8th International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment and Management, HNICEM 2015, 10.1109/HNICEM.2015.7393208

- Short Summary of Abstract:

In this article a method is described equipping an self-driving golf cart with LED’s to convey information to nearby pedestrians. By equipping autonomous vehicles with a feature like this, their performance as social robots is improved by building trust and engagement with interacting pedestrians.

  • Hussein, A., García, F., Armingol, J. M., & Olaverri-Monreal, C. (2016). P2V and V2P communication for pedestrian warning on the basis of autonomous vehicles. Paper presented at the IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC, 2034-2039. 10.1109/ITSC.2016.7795885

- Short Summary of Abstract:

In this article a method is discussed to broadcast positions from vehicles nearby to other road users and vice versa to minimize potential dangers and increase the acceptance of autonomous cars on roads.

  • Zimmermann, R., & Wettach, R. (2017). First step into visceral interaction with autonomous vehicles. Paper presented at the AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Proceedings, 58-64. 10.1145/3122986.3122988

- Short Summary of Abstract:

In this article the need of communication between pedestrians and vehicles is explored and if it could be achieved through motion behaviour of that vehicle

  • Florentine, E., Ang, M. A., Pendleton, S. D., Andersen, H., & Ang, M. H., Jr. (2016). Pedestrian notification methods in autonomous vehicles for multi-class mobility-on-demand service. Paper presented at the HAI 2016 - Proceedings of the 4th International Conference on Human Agent Interaction, 387-392. 10.1145/2974804.2974833

- Short Summary of Abstract:

In this article methods are described of conveying information and motion intention of autonomous vehicles to the surrounding environment.

  • Vasic, M., Billar, A. (2013). Safety issues in human-robot interactions. Proceedings - IEEE International Conference on Robotics and Automation 6630576, pp. 197-204

- Short Summary of Abstract:

In this article the safety in human-robot interaction is considered. First in industrial settings than with autonomous mobile robots operating in crowded environments (the most interesting part for us) and last with assistive robots.

  • Le, H., Pham, T.L., Meixner, G. (2017). A concept for a virtual reality driving simulation in combination with a real car. AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Adjunct Proceedings pp. 77-82.

- Short Summary of Abstract:

Human-machine interaction for autonomous driving is still under development. This article is about the area of increasing the level of immersion of virtual reality driving simulation with a real car.

  • Hacohen, S., Shvalb, N., Shoval, S. (2018). Dynamic model for pedestrian crossing in congested traffic based on probabilistic navigation function. Transportation Research Part C: Emerging Technologies 86, pp. 78-96.

- Short Summary of Abstract:

Pedestrians construct a virtual risk map that assigns the entire crossing area with probabilities for a collision with vehicles, and then select their actions based on their perceived probability for collision. A model is made which can serve as a standard tool in simulations for assessing accident risks in urban environments.

  • Dominguez-Sanchez, A., Cazorla, M., Orts-Escolano, S. (2017). Pedestrian Movement Direction Recognition Using Convolutional Neural Networks. IEEE Transactions on Intelligent Transportation Systems 18(12),8006277, pp. 3540-3548.

- Short Summary of Abstract:

This paper focuses on deep learning techniques such as convolutional neural networks (CNN) to achieve a reliable detection of pedestrians moving in a particular direction. This link might be useful if we have to work with convolutional neural networks: http://cs231n.github.io/convolutional-networks/

  • Deb, S., Strawderman, L., Carruth, D.W., (...), Smith, B., Garrison, T.M. (2017). Development and validation of a questionnaire to assess pedestrian receptivity toward fully autonomous vehicles. Transportation Research Part C: Emerging Technologies 84, pp. 178-195.

- Short Summary of Abstract:

This study analyzes pedestrian receptivity toward fully autonomous vehicles (FAVs) by developing and validating a pedestrian receptivity questionnaire for FAVs (PRQF).

  • Kim, T., Han, W., Kim, H., Park, Y. (2017). Vulnerable road user protection through intuitive visual cue on smartphones. CarSys 2017 - Proceedings of the 2nd ACM International Workshop on Smart, Autonomous, and Connected Vehicular Systems and Services, co-located with MobiCom 2017 pp. 13-17.

- Short Summary of Abstract:

This paper discusses how the most distracted road user type, i.e., smartphone users, can use the Basic Safety Messages (BSMs) from nearby vehicles to notice approaching danger and take appropriate defensive actions.

  • Dey, D., Martens, M., Eggen, B., Terken, J. (2017). The impact of vehicle appearance and vehicle behavior on pedestrian interaction with autonomous vehicles. AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Adjunct Proceedings pp. 158-162.

- Short Summary of Abstract:

In this paper, we present the preliminary results of a study that aims to investigate the role of an approaching vehicle's behavior and outer appearance in determining pedestrians' decisions while crossing a street.

  • Dey, D., Terken, J. (2017). Pedestrian interaction with vehicles: Roles of explicit and implicit communication. AutomotiveUI 2017 - 9th International ACM Conference on Automotive User Interfaces and Interactive Vehicular Applications, Proceedings pp. 109-113.

- Short Summary of Abstract:

This paper presents a study that aimed to identify the importance of eye contact and gestures between pedestrians and drivers.

Coaching Questions Week 1

What are you expecting to learn during the Robots course?

Learn to work together in an interdisciplinary group. Solving a problem while taking the USE aspects into account.

What kind of coaching do you expect?

We expect the coaches to correct us when we are heading into a wrong direction. Furthermore we expect them to motivate us for the subject.

What kind of coaching would you prefer?

We prefer a kind of coaching that does not simply tell us what to do, but asks questions that makes us think about aspects that we hadn't taken into account.

What will the coaches expect of you?

For us to ask questions about things we run into, instead of passivly wait for the coach to figure our issues out. Also, they expect us to seriously work on this project, and work as a team.

Week 2

Result of Meeting 1

From the meeting we learned that our problem is relevant enough, however our deliverables are still too vague. We need to have a very clear description of what we want to deliver, and why this is usefull. It also needs to include things that have not been done before. Also we learned that a master student is currently doing some research about our subject.

Who will do what?

This week, we collectively work on a new clear idea of what we want to deliver, and do some more state of the art research. Also, we will send the master student an email.

Deliverables

  • Write a report comparing different already existing solutions. The report will check the advantages and disadvantages of these solutions. Also, improvements (and possibly new idea's) of the solutions will be discussed.
  • Create a prototype of the solution that we think is best, and hasn't been created before. Also here there is room for making possible improvements.

Research Results

  • Existing idea's for car-human interaction

- Putting eyes on the vehicle that look at the pedestrian.

- Turning the windshield into a screen that displays the vehicles' intentions.

- Creating a face at the front of the car and use the side mirrors to display gestures to the pedestrian.

- Put an actual robot in the driver seat.

Meeting with Koen Roorda

Thursday morning 1st of March, two of us had a meeting with Koen. We first talked about what he did and then concluded that there is a lot to be done, however with the little time that we have we can't do a lot. We thought of both subjects, as well as deliverables. The subjects we came up with are

  • Using a chair-suit to simulate autonomous vehicles in the real world and investigate whether pedestrians would cross the road when there is a actual driver or a autonomous driver. This would be tested using various driving patterns.
  • Investigate how unwritten social traffic rules would be translated to a world with autonomous vehicles. In particular, will pedestrians take advantage of autonomous vehicles behaviour? So knowing that a vehicles has to stop when a pedestrian is crossing, will the pedestrian always cross?
  • Use webcams to gather data about pedestrians and car behaviour.

The deliverables we came up with

  • start of an research
  • literature study
  • Questionnaire framework
  • Short investigation (small videos)
  • VR simulation (very difficult and time consuming)

We now want to ask the tutors which of the deliverables would be sufficient. After the meeting this week we will decide on the final subject.

Coaching Questions Week 2

  • What is the most interesting thing you learned in the coaching meeting of the previous week and why?

We learned that our problem description is still a bit too vague. We need to be more concrete in what we want to research, and in the end delivered. Also we learned a master student is working on the same problem.

  • How did you incorporate coaches' feedback of the previous meeting in your project?

We sat down together to think about the direction we want to go in. This way we made our problem description clearer.

  • What new activities did you undertake during this week? What did you learn from these activities?

We did some more state of the art research to get a better understanding on what has been done and what subjects are still available. This way we got a better idea of what we want to deliver. We also sent the master student an email.

  • What did you do to prepare for next week's meeting?

We had a meeting ourselves to brainstorm about some new idea's.

Week 3

Result of meeting 2

We presented the idea's we came up with after talking to Koen to the coaches. After some discussion, the outcome of discussion is that we are going to tackle a new problem. This problem is the problem of people abusing autonomous vehicles. When people learn that autonomous vehicles will always stop, they will just pass the road without thinking twice. We want to research how people will act in a situation like this, and come up with a solution for it.

Problem description

Imagine a world in which autonomous vehicles fill the streets. For a lot of people this would be ideal traffic control. However, no human control over the car brings some setbacks. One of these is that people will abuse the autonomous vehicles, to their own advantage. An autonomous vehicle is programmed to stop for anything that comes in its way. This means that a human can just walk in front of the car quickly, since it will stop anyway. This leads to people mindlessly crossing the street, thinking the vehicle can't hit them. This is dangerous in many different ways. Ofcourse it is dangerous for the pedestrian, since the possibility exists that the car won't stop in time. It's also dangerous for the person inside the vehicle, since the vehicle has to stop very quickly, so the person inside will also feel this blow. Lastly, it's dangerous for other vehicles. There is a possibility that a vehicle behind the vehicle that has to hit the brakes is unable to stop in time, leading to a collision. To avoid these situations, action must be undertaken.

Users

The system will have a set of users, each with their own requirements. These might vary among the different user types. The pedestrians are a group of users that will indirectly make use of the system, by interacting with the autonomous vehicle which has incorporated this system. These pedestrians are concerned about safety and want to trust that this interaction does not fail. When wanting to cross the road, they should not have to perform actions that are too complex. Therefore the system should be easy to work with (ease-of-use).

The driver of the autonomous vehicle (or rather, the passenger) also wants to be able to trust this system as well as the autonomous vehicle itself. Since we target fully autonomous vehicles and not vehicles that still require some control of the driver, we envision the passengers of such a vehicle to trust what the vehicle is doing. The driver is also concerned with safety, accidents are to be avoided of course. Our system should be able to deal with all necessary interaction between the vehicle and the pedestrian, therefore the driver might not have to be involved in this interaction. We will have to determine whether this is the case when designing our system. The autonomous vehicle itself also counts as a user (even though it is not human). The workings of these vehicles should be improved with our solution and in traffic (autonomous) vehicle-pedestrian interaction should be safer.

Society

Our society should benefit from our solution. Governments spend millions of dollars already to increase traffic safety. Although we are definitely not at the stage where everyone drives an autonomous vehicle, we expect this to be the future and safety is always a concern when it comes to traffic. What we have to research is where exactly our system will be a solution to the problem. Pedestrian-dense neighborhoods where people are used to crossing roads with little care might require a different tactic then places where there are a lot of pedestrian crossings which are properly used. It is also taken to account how some people react to approaching cars. Some wait for contact and others just carelessly cross the road without any interaction/contact. Some Autonomous vehicles should be risk-averse and thus might be too careful when driving in environments like the center of Amsterdam, especially when we design a system that requires the vehicle to interact with each and every person that wants to cross the road.

Enterprise

Business-wise the system should be successful in that car production companies can buy and use it in their autonomous vehicles or that it could be installed at intersections or crossings. To get autonomous vehicles more accepted by the public, they have to become safer so that people can trust them. Our system might have a positive effect on this, showing to the public that these autonomous vehicles are safe.

Approach

We start by acquiring a setup. One of the ideas is a VR device, to simulate an autonomous vehicle riding towards the pedestrian. We can then observe how the pedestrian behaves in this situation. We can do this experiment with the same person multiple times, to see if their behaviour changes. Another idea is the so called 'chairsuit'. This is, like the name says, a suit that makes a person look like a chair. When a person in the driver seat uses this suit, it seems like there is no driver in the car. We can use this to drive around a pedestrian crossing and see how people react. After this, we can ask them some questions. The next thing is to analyse the data we acquired by one of the methods stated above. After this, we can try to come up with a solution for the problem.

Deliverables

A report about the results of our experiment, and possible improvements based on these results.

Planning

Plan A (VR):

Week 4:

  • Mail Koen Roorda about the VR hardware and software (+permission to use, time/place to use).
  • Get familiar with VR ourselves (+ write what kind of experiments we want to do and how we are going to do it, write a program for the VR system).
  • Find test subjects to conduct experiments with.
  • Write experimental setup(+Introduction), define USE aspects (and summarize few articles for more background info).

Week 5:

  • Conduct the experiments and gather data
  • Analyze the retrieved data

Week 6:

  • Write about our results of the experiments
  • Write conclusion
  • Prepare the presentation

Week 7:

  • Finalize the report for the deadline


Plan B [Chair Suit]:

Week 4:

  • Mail to Jacques Terken about the chair suit: Jorik
  • Start on the report, write experimental set up and define the USE aspects for this specific topic: Jeroen & Tim
  • Write summaries about all the state of the art articles we have read, so we can easily refer to them: Everybody about their own articles
  • Make sure to have a car ready: Lisanne

Week 5:

  • Create and edit the video's
  • Meeting on monday
  • Analyze the experiment
  • Write report sections about the experiment results

Week 6:

  • Start the discussion/conclusion
  • Start to work on a presentation
  • Write recommendations

Week 7:

  • Finish the report

VR plan

Introduction

The technology of fully autonomous vehicles is on the rise, as well as the production of these FAV’s. They will have to be able to drive safely from any point A, to any point B. Not only should they handle dangers that exist currently (in a society without FAV’s), but they should also be able to deal with new dangers that are specific to autonomous traffic.

One of these dangers lies among pedestrians. As it stands currently, pedestrians that want to cross the road adhere to some social rules and can perform some communication with the driver to safely make the crossing. Although FAV’s might have some substitute for this communication (see papers), they will still take the safer route if possible, which in the case of pedestrians crossing is to stop if the situation gets dangerous.

The follow-up question is whether pedestrians might misuse this programmed behavior of FAV’s to cross the road quicker, as they are learning the behavior of FAV’s. This misuse might also result in passengers of FAV’s feeling less safe, since they might be aware that there are pedestrians that will just walk in front of the car since they know it will stop for them.

As we are looking into this problem of FAV misuse when crossing the road, we will also think of some possible solutions. However, our main focus is to list ways in which people will misuse FAV behavior so that future research can be done into solutions to these specific types of misuse.

Experiment plan

Our problem is only found in a society where FAV’s are the norm and where on roads there are little to none human-driven cars found. Sadly we do not live in such an environment (yet), thus we will make use of virtual reality. In virtual reality we will build a traffic environment wherein the user of the headset can walk around as a pedestrian and cross roads on which FAV’s are driving. In our experiment, the user will have to walk from a point A to a point B. Subjects

The pedestrian that we would have in mind for this experiment is one with a lot of haste towards his or her destination. This pedestrian should also have a general feeling of how an FAV will react to their actions, since in a world with FAV’s almost all pedestrians will have this understanding. To obtain these two assumptions we do the following:

  • We tell the user to get from point A to point B as fast as possible. This will incur some haste into the user, prompting them to take the quickest option that they deem safe. This will be in line with day-to-day pedestrians wanting to get somewhere quick, taking actions they would not do when not in a hurry.
  • We give the user some amount of attempts (TO DO: 10 tries?). This will result in the user understanding the behavior of the FAV’s more in the last few attempts. Especially the way the vehicles react to their own behavior should be learned. This will then be in line with how pedestrians in a society with FAV’s have a feeling for their behavior.
  • During these 10 tries we will not tell the user to ‘improve their time’. We want to step away from any game-like aspects, as this will prompt the user to take game-like actions, abusing not only the FAV’s behavior but also abusing the fact that it is a simulation. More importantly, the times recorded will not be used at all.
  • Lastly, to make sure the user takes the simulation seriously, getting hit will result in them not getting any more tries. As people who have been in a traffic accident in real life are more careful afterwards, these people will have no use for the experiment anymore. This is because they now know what it is like to have been hit without repercussions, resulting in a more game-like feel to the experiment.

Environment Simulation

The simulated environment will have some requirements. Firstly, the environment should not look too unrealistic or cartoonish. This will result in the subjects feeling it is game-like. Secondly, in this environment the route from A to B should be clear. We want the subjects to know where to go. We do not want them to get lost on the way, especially since they are told to get to point B as fast as they can. Lastly, the route should contain a set number of crossings (note however that the environment should be the same for each subject). Ideally, there is only one route from A to B that will have this number of crossings.

Each crossing should then have one or more FAV’s driving on them. To generalize this experiment a bit more, we can give each crossing a certain traffic density value. In other words, one crossing might have as much FAV’s driving on it as we might find in a large city like Amsterdam, while another crossing might be similar to one found in rural areas where the subject might only see one car pass by.

Analyzing results

Now we perform the experiment, using X subjects. During each attempt by each user, we monitor the behavior of the user and look at how they acted in this traffic simulation as a pedestrian. To do this the software has to be able to record their attempt so we can analyze the footage afterwards. A simple way to do this is to have the computer monitor present the view of the user and recording the screen footage with third-party software (OBS).

The attempts we are most interested in are the last few of each subject. This is because they are then the most used to FAV’s. However, we will have to look at all attempts to see whether a user did actually learn the FAV behavior. If the way the user crossed roads from A to B does not differ much between the first and the last attempt, they either did not misuse the FAV’s at all, or they misused the FAV’s from the start.

When we have collected footage of subjects who got used to the behavior of the FAV’s and misused this behavior, we will also be able to think about solutions to these types of misuse.

Seat Suit plan

Introduction

The twentieth century has brought the world one of the most widely used passenger transport technologies, namely cars. They allowed people to drive freely to wherever they wanted to go. Today, a society without cars is almost unthinkable. However, companies want to push cars to the next level. That level is cars being fully autonomous. Autonomous vehicles can drive from point A to point B without the need of a person controlling the vehicle. By removing the need for a driver, they are expected to reduce the amount of collisions resulting from human errors on the road. This will greatly improve safety in traffic. It will take years before autonomous cars are fully integrated into traffic, but the first steps are already being made. Some automotive companies have already integrated self-driving features in their design, for example Tesla’s ‘Autopilot’ and Volvo’s ‘Pilot Assist’. These are the first steps to vehicles being fully autonomous. However, there are already some self-driving vehicles on the road today. The world’s first autonomous taxi service was launched in Singapore in 2016. A couple months after this, self-driving busses took to the road. A world with self-driving vehicles is surely on the rise. Research regarding the technology of autonomous vehicles has increased dramatically in recent years. However, only a limited number of investigations regarding interaction between pedestrians and self-driving cars have been conducted. Knowledge on this field is just as important as creating the vehicles themselves.

In a recent study from February 2018, about 1000 participants were asked their opinion about autonomous vehicles, particularly the safety of autonomous vehicles. Their opinions depended greatly on their road-user perspective. As pedestrian, the participants felt safe, but as passenger in the vehicle, some concern existed. There were also some concerns about crossing a street. Currently, pedestrians that want to cross the road can communicate with the car driver, for example in the form of eye contact. This communication has become a social rule in traffic that people follow unconsciously. However, in a world where vehicles are autonomous, this social aspect is taken away. Pedestrians can no longer make contact with the driver, which is why they express concern for the matter. There are some solutions to this problem, for example displaying the car’s intentions on the windshield. However, most of these solutions haven’t been tested yet.

The goal of this research is to find out how people react to a driverless car when crossing the streets. This will be done by an experiment with a car which appears to have no driver. This is done by equipping a so called ‘chairsuit’. The method of the experiment will be explained later. From this experiment we hope to gather data of how people behave in the situation illustrated above. The data could possibly help with finding a solution regarding the pedestrian-vehicle communication problem.


Experimental Setup

The experiment has to simulate a world in which all vehicle are autonomous. This is done by using a seat suit which covers the actual driver of the vehicle and makes it look like a driverless vehicle. The goal of this experiment is to investigate which solution, if any, will work best in ensuring the pedestrian is safe to cross the road. There are two ways to go about this experiment. If there is a lot of time, prototypes can be made of the vehicles and the solution. Then the experiment could be conducted in real time with actual pedestrians. Since there wasn’t much time the other method was chosen. The vehicle was filmed from a pedestrians point of view that wants to cross. In post-production the solutions were visualized on the vehicle using computer software. These videos are then shown to a group of people who need to say when they feel it is safe to cross.

A few assumption and choices were made for this experiment. The choices are made in such a way that an environment is created where the pedestrians do not have the right of way by default.

  • There is no indicated crossing
  • There are no traffic lights
  • There are no signs
  • The road is a standard 2 lane road

Also, in order not to influence the decisions of the participants based on the appearance of the car, a normal looking car has been used.

There are 3 adaptions made to the vehicle to make it communicate with the pedestrians. The control group in this experiment is an ‘autonomous’ vehicle without any further adaptations, autonomous in the sense of a car suit. The first adaptation is a light on the car when the pedestrian is safe to cross the street. The second adaptation is an arrow which is projected on the road, this shows the pedestrians that they can cross safely. The last adaptation to the vehicle is a pedestrian crossing which is projected onto the road when it is safe to cross for the pedestrians.

The videos are shown to people via a questionnaire. They will have to push a button when they think it is safe to cross. In this way data is gathered which solution has the best impact on the feeling of safety of the pedestrian. One important detail is that the people who fill in the questionnaire do not know up front what the adaptations to the car mean.

Coaching Questions Week 3

  • What is the most interesting thing you learned in the coaching meeting of the previous week and why?

We finally got a good idea of what we actually want to research, and what we want to achieve. Also, we learned that we don't have to focus on achievability too much, but rather focus on what we want to research, and make that achievable.

  • How did you incorporate coaches' feedback of the previous meeting in your project?

We left the achievability out of the problem some more. So instead of doing the experiment a lot of times, we will scale down to just a few times.

  • What new activities did you undertake during this week? What did you learn from these activities?

We started working on acquiring the needed things to conduct an experiment. Also we started with wrtiting the report.

  • What did you do to prepare for next week's meeting?

We had a meeting again ourselves.

Week 4

Result meeting 3

We told the coaches clearly what we wanted to research, and they agreed with our plan. However, we learned that we have to divide the work some more and make a clear hypothesis.

Plan for this week

This week we are going to make the video's we want to show our participants. We also want to make a clear hypothesis, and complete a questionnaire. We'll also continue working on the final report.

Filming

On Monday morning the movies were recorded on the TU/e. One person with the suit seat on drove through a street and made a stop at the camera. The camera was place from the pedestrians point of view to make sure the test persons can imagine the best what they would do in the situations. There was also made sure there was a crosswalk at both sides of the road. The car drove several times through the street without anything on it for the reference movie and for the movie with the text on the window and the eyes. These two solutions were to hard to put on a real car and will be edited in the video later. The light was able to be put on the car and also several movies where the car drove to the street and stopped before the camera (place of the pedestrian) were recorded.

Hypothesis

It is expected that people will feel more safe with the solution on the car. So they will cross the street earlier, when the autonomous car can show to the pedestrian that it has seen them and let them know they can cross the street. From the many solutions there already are three will be tested: a light on the car which goes on when the pedestrian can cross the street, text that will be displayed on the car window and eyes that follow the pedestrian like a driver would do. Expected is that the solution with the light will work best. This because it looks a bit like a traffic light and people are already familiar with this. It will probably also be the cheapest solution to implement. For now only one light will be put on the car but this can also extended later to two or three lights to make it look more like a traffic light. The text on the window will also be clear but because person has to read it before it can know what to do it is expected to take longer for the pedestrian to know it is safe to cross the road. It is also completely new to pedestrians, so later when people already know what the car will display when it is safe to cross the road it will take less time. Last the eyes, from this solution it is expected that it works the worst. This because it might be unclear for pedestrians what to do when the eyes look at them or not. The eyes can not give any expressions like a driver can so the pedestrian might not know what the car means and what to do.

Questions questionnaire

  • Age/gender (Get some general information about the test persons)
  • When does it feel safe to cross the road? (During the movie people have to push a button when they feel safe to cross the road) (every movie)
  • What made you decide to cross the road? (Open question) (every movie)
  • Which solution showed in the movies was the clearest way for you to know that you could cross the road? (Multiple choice: light, text, eyes)
  • Was it clear for you in the first movie when you could cross the road safely? (yes/no)
  • If yes, what made it clear? (Open question)
  • Was it clear for you in the second movie that it was safe to cross the road when the light lit up? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Was it clear for you in the third movie that is was safe to cross the road when the text was displayed on the car window? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Was it clear for you in the fourth movie that it was safe to cross the road when the eyes of the car looked at you? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Do you have any other suggestions to make it clear for pedestrians to know when it is safe to cross a road when an autonomous car is approaching? (Open question)

Plans next week

For next week we want to finish the video's and the questionnaire so we can get results from the questionnaire. We also want to make a start with analyzing these results if we get them in time and work further on the report/paper.

Coaching Questions Week 4

  • What is the most interesting thing you learned in the coaching meeting of the previous week and why?

We have to still be more clear in our hypothesis, and have a task division.

  • How did you incorporate coaches' feedback of the previous meeting in your project?

We worked on being more clear. We tried to make it more clear what we did this week and what we are going to do next week.

  • What new activities did you undertake during this week? What did you learn from these activities?

We shot the video's of the car approaching. We filmed some of the car itself, and some of the car with an extra light on it. We also worked on a hypothesis for our research, and came up with some questions we can ask our participants. Lastly, we tried to install and become acquinted with photoshop software, to create our video's.

  • What did you do to prepare for next week's meeting?

We had a meeting ourselves, and put all our findings online.

Week 5

Results Meeting 4

The video's are well made, the only thing we still have to do is think about different ways to ask our questions to the participants, create the questionaire, and analyse the results.

Plan for this week

Edit the video's so that our solutions are visible on the car. It should look as real as possible. Also, think about questions, create a questionaire and send it out to people, so we can start gathering results.

Video's

Link to the corresponding video's can be found below.

  • The video of the vehicle without one of the solutions on it. [1]
  • The video of the vehicle which shines a green light when the pedestrian has been spotted. [2]
  • The video of the vehicle which shines a pedestrian crossing on the ground once the pedestrian has been spotted. [3]
  • The video of the vehicle which shines a green arrow on the ground once the pedestrian has been spotted. [4]

Questionnaire

Introduction

During this questionnaire you will be shown a couple of clips. The situation is as follows: You are a pedestrian and live in a time where the switch from normal vehicles to only fully autonomous vehicles has just been made. You want to cross a road without a pedestrian crossing. There is a vehicle coming your way which is driving at a speed and distance such that the car first needs to stop or pass by before you can cross the road. In every clip you are asked to indicate when you think it is safe to cross the road. You can indicate this by pausing the video and fill in the time shown in the bottom right corner of the clip. After all clips have been shown a few questions about your perception will be asked. Every clip takes about 10 seconds and you will be shown 8 clips. The total time for this questionnaire will be about 5 minutes.

  • Age/gender
  • At what moment do you cross the road?/ Push the button when you would cross the road. (During the movie people have to push a button when they feel safe to cross the road) (every movie)
  • What made you decide to cross the road? (Open question) (every movie)
  • Was it clear for you in the first movie when you could cross the road? (yes/no)
  • If yes, what made it clear? (Open question)
  • Was it clear for you in the second movie that it was safe to cross the road when the light lit up? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Was it clear for you in the third movie that is was safe to cross the road when the arrow was displayed? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Was it clear for you in the fourth movie that it was safe to cross the road when the pedestrian crossing was displayed? (yes/no)
  • If no, what would you change to make it clear? (Open question)
  • Which solution showed in the movies was the clearest way for you to know that you could cross the road? (Multiple choice: light, arrow, pedestrian crosswalk)
  • Do you have any other suggestions to make it more clear for pedestrians to know when it is safe to cross a road when an autonomous car is approaching? (Open question)

Possible improvements:

  • instead of if yes/no -> yes/no, could you give an explanation?
  • Testing before gathering real participants
  • Rank scale (rate between 1-5)? -> was it clear scale

Coaching Questions Week 5

  • What is the most interesting thing you learned in the coaching meeting of the previous week and why?

We need to think about different ways to ask questions to our participants. The way the question is asked makes a difference in the respons of the participant.

  • How did you incorporate coaches' feedback of the previous meeting in your project?

We adepted our questions.

  • What new activities did you undertake during this week? What did you learn from these activities?

We editted the video's, made some questions and sent out a questionaire.

  • What did you do to prepare for next week's meeting?

We had a meeting ourselves again.

Week 6

Results meeting 5

We presented the questionaire and the video's we made to the coaches. We learned that it may be better to show each video multiple times, and in random order. Also, that we will have a lot of work and little to to finish the research.

Times of first encounter with solution
Times of second encounter with solution

Preliminary Results

These are preliminary results based on the response of 40 participants. The results will be discussed more expicitly in the final report.

Video Results

The results of the video's are shown in the graphs on the right. The top graph corresponds to a scatter plot of the time it took participants to cross the road after they saw the video for the first time. The middle graph shows the same thing, but after the second time they saw the video with corresponding solution. The average times it took our participant's to cross the road can be seen in the bottom graph. The graph shows the mean time for the first time they saw the video (black) and the second time they saw the video (red). It can be seen clearly that the solutions do work, and people feel safe to cross the road earlier. The arrow and pedestrian crossing seem to work best. The biggest anomaly here is the difference in time between the first and second time participants see the video of the car with the light on it. The second time they saw the video they responded a lot quicker. This will be explained by the open questions in the next section.

Results Open Questions

  • No Solution

The participants indicated that they crossed the road after seeing the car break, which is of course to be expected.

  • Light

For the first solution participants also indicated that they waited for the car to break. This indicates that the solution wasn't clear enough for them at first. When they were asked to explain, most participants mentioned that they weren't sure whether the light was meant for them, or maybe for another car. Some also felt it to be more of a warning sign than a safe sign. This also explains the big difference in times the first time they saw the video and the second time they saw the video. The first time they weren't sure if it was meant for them, so they waited a bit longer. However, from this they learned that the light was actually for them. Therefore, the second time they saw the video they crossed the road quicker.

  • Pedestrian Crossing

Accoording to the participants, this solution was very clear. Everyone crossed the road due to the pedestrian crossing being highlighted on the ground for them. When asked why, they replied that a pedestrian crossing corresponds to safety, which gave them the feeling they could cross the road safely. However, participants felt that the solution could be improved by shining the pedestrian crossing in a steady location where the pedestrian can cross, instead of shining it in front of the moving car.

  • Arrow

This solution gave some differences between participants. Some understood it, saying the green colour made it feel safe, and the direction of the arrow told them they could cross. On the other hand, some participants felt that the arrow wasn't visible enough from a distance, so waited for the car to come closer before they crossed the road.

Coaching Questions Week 6

  • What is the most interesting thing you learned in the coaching meeting of the previous week and why?

We learned that it will be good to show the video's multiple times and in random order in our questionaire.

  • How did you incorporate coaches' feedback of the previous meeting in your project?

We incorporated these things in our questionaire.

  • What new activities did you undertake during this week? What did you learn from these activities?

We sent out our questionnaire to people, and did a preliminary analysis of the data. Also we prepared for the presentation.

  • What did you do to prepare for next week's meeting?

We had a meeting ourselves again.

Week 7

Final Report

A link to our final research report can be found here.Final Report

Coaching Questions Week 7

  • What are the major steps of the project? Please list
    • State of the Art research
    • Defining the problem which we wanted to solve (This took way more time than expected)
    • Creating the experiment (Also took more time as expected)
    • Conducting the experiment
    • Analyze the results
    • Write the report and figure out what we could have done better.
  • What is the most important thing you learned in this project? (e.g .about design or working in groups, etc)

Working in an interdisciplinary group, where everyone has different visions. It was difficult to get on the same line together, therefore the project did not go as fast as we had hoped. Since the ideas within a person were not always the same as the ideas within the group. In the end we got through to each other, so we could actually work on the same project together. No one of us has actually done such a research before. This is also the reason why not everything was taken into account when conducting the experiments. The result of this is that we actually investigated something else than we actually wanted. From this we learned that you really need a clear view, take a step back and really test the question you are asking before you conduct the actual experiment. Sadly our lack of knowledge made it so that we did not research the thing we wanted to, but at least we learned a lot from our mistakes.

  • What do you wish you had spent more time on or done differently?

Because of the limited time, we had to resort to creating video's for our participants. If we had more time, we would like to use virtual reality instead of the video's, so participants get more immersed in the situation. Also, we only made video's with the same driving style and traffic. However, it would be nice to be able to compare the results we gathered with results of video's (or VR) on bussy roads, or where the car has different braking patterns, to see how this effects the participant's choice and to eliminate various learning processes which could influence the results.

  • What was the most enjoyable part of this project? Please explain why

The most enjoyable part was when we could actually start the research. Of course having a good problem description is key, but when we could actually go out and do what we wanted to do, the project started becoming a lot more enjoyable.

  • What was the least enjoyable part of this project? Please explain why

This was probably when we got stuck on what we actually wanted to research, so we had to gather even more articles and information. However in the end, it all paid off! Also, working in the wiki was not a good experience for us, since the program is not really useful and fairly limited.




Coaching Questions Group 16

Personal tools