PRE2018 3 Group14

Group Members

Name	Study	Student ID
Joost de Boer	Software Science	1016745
Yanic Dobler	Software Science	1007100
Leon Kersten	Software Science	1006966
Pietro Maschera	Psychology and Technology	1220953
Koen Vlaswinkel	Software Science	1016271

Problem statement

Traffic management and prediction is a major concern and industry all over the world. City planers are always on the lookout for new methods, algorithms and procedures to guide the ever-growing car masses through the narrow streets of urban areas. These engineers focus mostly on developing means of controlling traffic light switching and while it is an important area to optimize, it has one looming downfall: It can only dictate the paste and rythm at which traffic flows through a specific intersection and not how traffic is split among possible intersections. This is where our product comes in, instead of pushing the state of the art on traffic light control, we aim to develop methodologies to direct single cars to take an alternative route with less traffic. This is a trivial problem when applied to a single car, but quite an algorithmic challenge when all traffic tries to find alternative routes in real time and they are able to communicate their current location to each other.

Objectives

• To evaluate the current state of the art on traffic advice.
• To evaluate the current state of the art of traffic light control based on actuators.
• To interpret real-life traffic data provided in real time to feed the simulation (below).
• To develop an algorithm to solve the aforementioned problem.
• To develop a traffic simulation showing the effects with and without the provided soltuion on heavy traffic in urban areas.

USE Aspects

Target users

Our project aims to provide motorized traffic (with and hopefully without drivers) with individual traffic advice, hence our main target users are traffic users. However, we aim to develop an algorithm and a simulation to show the effect of the algorithm, not a user friendly interface to make the algorithm usable by the average car user. Hence self-driving car manufacturers, city planners and traffic management engineers are also part of our target users as they ought to profit from the algorithm and simulation software when solving their own challenges.

What do they require

In general our target audience needs a system which allows their car to calculate individualized traffic advice in their current environment with respect to surrounding cars and known road layout (& rules) but also considering unknown traffic light behavior given the ever changing shift in traffic as each car separately tries to find the optimal route.

Users

Users would not be using this software directly, but they would be using it passively. In other words, car drivers would not be inputting any data nor would need to simulate the outcome of the data, but enterprises and engineers could use it to build software's that could profit the average car user with ways to avoid traffic jams and choose better routes.

Society

Society could benefit from the entrance of this software in the market. The private car owners, once they bought a product that has been engineered on the base of this software, could have a smarter and better experience with their car travels. This could benefit the population as a whole, increasing the viability in streets and introducing a enhanced management of traffic jams. The result of these improvements could even have effects in sustainability and in car pollution. Moreover by trying to simulate different scenarios of road branching and understanding what could work better, governments could use this software to build or re-engineer better and smarter roads, once again to improve viability and pollution.

Enterprise

Enterprises could take a lot of advantages from this software. Examples of enterprises that could benefit from this service could be autonomous cars or sat navigators manufacturers. By inputting data from cities or highly frequented roads these companies could simulate and then decide, on the base of the simulation, which roads are better to be taken from an autonomous car or from a car user.

Approach

Our approach will be based on mathematical and logical models with whom we aim to describe the traffic behavior around the vehicle and also derive the best possible advice for the individual vehicle. To do so we begin by understanding current models for traffic behavior and prediction. We then use this knowledge to formulate new models which fulfill our purpose. Using these models and real life traffic data a simulation is created, showing traffic with and without the created algorithm. The simulation will "probably" be made using openGL and java.

Planning

In short, we want to do a research on the topic of flow control in traffic situations and improve the flow by analyzing data and creating a software model. The research will be detailed further on the wiki.

There are two things that keep occuring each week, one before each meeting, and one at the end of the week. What is done at the start of each meeting is discussing work done that has been divided last meeting. What is done at the end of the week is a short evaluation meeting with the tutor to compare progress and discuss issues that might have come up. This will start in week 2, as week 1 is the introductory week, with no assigned meeting.

• Week 1:
  • Brainstorm on the project idea and pick one idea.
  • Elaborate on it by finding sources(indiviually, at least 5 per person).
  • Divide the work for the first meeting and work on this individually. Make sure all of the points of the slides have been covered before the first meeting with the tutor.
• Week 2:
  • Start researching with the found papers, looking for how flow control works currently and finding ways to help improve it.
  • Look for databases of traffic data (in the Netherlands, or outside of it), which might be used in later research/in the software model.
  • Search for algorithms or discuss and make one on our own.
• Week 3:
  • Start work on a software model based on an algorithm found/created in week 2. Collectively implement a start for the algorithm using data.
  • Keep on researching for databases to use and finalize on a choice for database to use in the software model.
  • Write about the research that has been done in week 2.
• Week 4:
  • Implement details of the software model and fix possible bugs. Discuss ideas that might improve the model, and implement the data reading such that we can get results on our improvement.
  • Continue working on writing about the research done, and report on the software model on the wiki.
  • Research for new studies/algorithms should not be needed after this point, as the software should be largely done.
• Week 5:
  • Finalize the software model and process data from the database. If needed, fix bugs in the software model.
  • Compare the data from the results and think of conclusions linked to the outcomes.
  • Collectively look for reasons why the data behaves like how it is found.
  • If there is time for it, start working on the questionnaire for the USE aspect.
• Week 6:
  • Continue processing of the data and results of using the software model. The software model should be done at this point, because otherwise data will not be reliable for later.
  • Find interesting results, and compare them to real life data. Couple the found results to USE: is it helpful in improving daily life? Done in the form of a questionnaire perhaps.
  • Start with picking interesting topics to discuss about our results for the presentation in week 8.
• Week 7:
  • Finalize a presentation about the work that has been done on the project.
  • Finish conclusions on the results, and update the wiki accordingly. If it did not turn out to improve flow, explain why. Similarly, if it did improve flow, show how this was done.
• Week 8:
  • Give the presentation for the tutors.
  • Update the wiki with last information on the project.

Who does what

The general planning per person:

• Koen: Software model.
• Yanic: Database Analysis.
• Pietro: Research of the USE aspect.
• Leon: Theoretical Analysis of Algorithms.
• Joost: Data Analysis and Research of the Flow problem.

This planning will be updated weekly to include what everyone will have to do for the following week.

As of 11/2/2019, week 2:

• researching papers: Yanic, Pietro
• looking for databases: Koen
• searching for algorithms: Joost, Leon

Milestones

• Evaluation of both main topics completed
• Complete report on opportunities and threats of their co-operation.
• Interpret real life traffic data to feed the simulation
• Develop functioning algorithm for the problem.
• Having a simulation prototype making use of real life traffic data.
• Combining everything together to create a working simulation showing the positive effects of the algorithm on urban traffic.

Deliverables

• Research and literature study
• A flow control algorithm
• A model for simulating and testing our algorithm
• A functional simulation
• An analysis of the feasibility
• This wiki page on all our findings and processes

SotA

Previous Groups

Subject and relevancy	Summary	Group
Vehicle intersection control This could be interesting to see what final algorithm they have introduced to solve the problem. They only work on autonomous cars, even if a part of the algorithm is dedicated to traffic lights in order to have the best solution in an environment of both normal and autonomous cars. Moreover their project was mainly for research sake - no actual simulation.	The idea is to design an algorithm that will use the benefits of autonomous cars on intersections with mixed traffic.	PRE2016 1 Groep2
Traffic light solution They have had kind of a similar idea to ours. They picked the ring of Eindhoven as a model and they used Netlogo as the software of simulation. They only focus on lights and green waves - no closing streets.	One of the methods to improve traffic flow is through the use of green waving. Here cars can continuously get a serie of green lights when they drive at an appropriate speed. For this project it was decided to try to implement green waving in a ring system. This was modelled through the use of the program Netlogo, where eventually a ring of twelve crossings was achieved.	PRE2016 3 Groep11
Pedestrian crossing This can be interesting for a research point of view and the method the used, even if it involves only autonomous cars and pedestrians. They conducted interviews, questionnaires and asked to specialists to gather information. They looked also at experiments made from the big companies (Tesla and Volvo).	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.	PRE2017 3 Groep16
Self-learning navigation software This can be cool for the study of traffic jams that they made and the actual navigation system that could be something collateral to our project.	Every day in the Netherlands people commute to work with their cars, and every day the same thing happens, traffic jams causing a lot of pollution and waiting time. Most of the time people are traveling a lot longer than necessary to reach their goal. This gave us the idea to create a system that will reroute this traffic through secondary roads to minimize the overall waiting time or maybe even prevent traffic jams. By doing a lot of user research we hope to create a user friendly system that will help the user and society with this problem.	PRE2016 3 Groep13

Traffic light SotA evaluation

The current industry standard on traffic light control varies a lot within areas, cities and countries. Most use a fixed, cyclic schedule which was determined by some software or a human engineer. Other, more sophisticated crossroads might employ a dynamic system which uses factors such as day of time, holidays and other trends to control the traffic light, but in essence it is still a system that repeats, at least partly. Current SotA research develops into two major directions:

• The development of traffic light control systems using sensors to determine traffic load on the relevant lanes. Based on the actuator data, the system would leave open busy lanes for a longer time to increase traffic flow and reduce congestion.
• The development of prediction systems using big data and artificial intelligence which aim at predicting, with a high accuracy, the traffic behavior of the future. Using these predictions the traffic lights would then adjust its cycle to maximize traffic flow and reduce congestion.

The first direction runs into a widespread issue of robotics, any type of sensor is vulnerable to noise and can occasionally provide false information. Distorted information would lead these systems to break and/or misbehave. Consequences could be at worst fatal. The second direction is more predictable and less error prone, but it remains to be seen whether traffic can truly be predicted with a high enough accuracy for the process of developing such an AI to be worth the resource investment necessary. Also any unforeseen change in traffic due to outside circumstances will render the prediction model useless.

There is one more type of approach which was found to be very interesting and potentially a great co-system for what we are developing. The idea is that each car, given a starting point and a destination, receives a virtual "budged" to perform its journey. Traffic lights all have spots in their lanes which incoming cars can "purchase" using previously mentioned virtual budged. A car can only enter the lane of a traffic light if it has purchased a spot for itself. Upon passing the traffic light, the spot can be sold again by the traffic light to another vehicle. The vehicles on-board software would constantly engage with the environment (traffic lights, toll gates etc.) in a virtual market place, trying to find the cheapest spots which allow for the quickest route. This system in particular is only believed to be implementable if self driving cars make up the majority of traffic, however the idea of being able to assign numerical values to routes based on demand is a valuable corner stone for our own algorithm, as it provides another numerical fix point which is determined by the system of traffic vehicles and not on a potentially falsified.

State of the Art Traffic Congestion

Most researchers agree that the most effective way to fix congestion in cities is implementing tolled roads. These tolled roads should be priced based on marginal cost (including marginal social cost to account for externalities). Vehicles that carry high number of passengers such as buses should have a reduced price or be exempted from the charging system, as these vehicles should be encouraged. This is because, if more people use those type of vehicles, less vehicles overall are needed, and hence these vehicles help reducing congestion. An example of a country which already implements this system quite effectively is Singapore where automatic charging gates charge cars automatically the moment it drives onto a tolled road.

To fix traffic congestion however, good traffic congestion detection is needed. At the status quo this is done by tracking a set of cars (usually taxis) with GPS systems, and then checking the positions of those cars occasionally. Shared Nearest Neighbor Algorithm is often used to find congestion after those GPS systems are implemented. observation.

Other relevant State of the Art papers

Tettamanti et al. discuss a method of responding to real-time traffic using a signal split algorithm taking uncertainty into account. They try to minimize the queue length while doing so.

Ge et al. look at how a traffic jam stabilizes and does both an analytical and simulation study. They derive that a three-car lookahead is enough for cooperative driving.

Konishi et al. use a single-lane model and a lead car and following vehicles to observe traffic jam formation. When the speed of the lead car is varied, there are varying degrees of traffic jams. They derive a condition under which no traffic jam is formed.

Davis adds reaction time (delay) to an already existing model to observe how driver delay impact the flow of traffic. In his modified model, traffic jams are caused primarily by driver reaction time.

Koukoumidis et al. show how mobile phones can be used to reduce delays caused by traffic lights. Users receive the schedule of traffic lights ahead on their mobile phone and can adjust their speed so as to avoid coming to a complete halt. They evaluate SignalGuru, a service that collects and predicts the traffic signal schedule.

Mellodge et al. discusses the details of a path following lateral controller implemented on a prototype car by using a kinematic model for a wheeled robot. The complete architecture of this is discussed and how it has been simulated.

Ge proposes a new car-following model where the optimal speed is determined based on the velocity of the two cars in front of it. They also discuss the steady and error states and how they occur, including a state in which no traffic jam occurs. Moreover, a control scheme under which no traffic jam occurs is discussed.

SUMO is an open source traffic simulation package including supporting tools. Krajzewicz et al. discusses the current state of the package as well as its major application and future developments. The current version is available as well. [1] Traffic Control Test Bed is an open-source piece of software built on SUMO that allows traffic control systems to be evaluated and benchmarked against each other.

MITSIM is another piece of software that was developed for modeling traffic networks, just like SUMO. It includes advanced traffic control, route guidance and surveillance systems. It simulates invididual vehicle movements. The current version is available as well. [2]

Another open-source software developed for traffic networks is YatSim. This software focuses on testing consensus-based control strategies in urban road networks. By using a consensus-based control strategy, the traffic system requires agreement to guarantee several factors.

A test bed for multi-agent control systems is presented by Van Katwijk et al. The test bed was created to aid research and two examples are discussed.

Machine-to-machine communications can be used for road traffic management as well, as discussed by Foschini et al. Their design is built on top of existing production systems, unlike other proposals at the time. A year later, another paper on this was published by Fernandes et al. They discuss mitigations for communication delays inherently present and simulations of intraplatoon information management strategies. They find that the effects of communication delays may be cancelled out by better communication.

Van Arem et al. discuss the impact of cooperative adaptive cruise control on traffic flow. Cooperative adaptive cruise control is an extension of adaptive cruise control which also communicates with the predecessor by wireless communication. This allows a shorter following distance and a simulation executed by the authors show an improvement of the stability of the traffic flow and a slight increase in the traffic-flow efficiency.

Papageorgiou et al. show how a reduced throughput can be achieved and a number of countering methods. They discuss these methods in three context: urban road networks, freeway networks and route guidance.

Database analysis

The NDW provides real-time, as well as historical traffic information. [3] [4]

OpenStreetMap provides map data that we can use for visualization. [5] [6] [7] [8]