An automated signalized junction controller that learns strategies from a human expert

Box, Simon; Waterson, Ben

doi:10.1016/j.engappai.2011.09.008

Cited by 13 publications

(13 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Fixed time control provides a benchmark against which human control performance can be compared it should be noted that most real world junctions use an adaptive control system such as MOVA [2] or SCOOT [3]. Unfortunately these adaptive controllers were not available for this test, although they have been tested against humans in simulation based computer games [4], [5].…”

Section: ) Control Methodsmentioning

confidence: 99%

“…The results of this research have indicated that players of the game can be very good junction controllers and even outperform some of the market leading automated junction control systems, such as MOVA [2] and SCOOT [3]. This particular result has prompted the development of machine learning junction control systems that can learn control strategies from human players of the game and from experience [4], [5]. Since obtaining the above results the authors have been keen to collect data on the performance of human controllers in "real-world" tests in a controlled environment.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Lessons from proving ground experiments to investigate junction control

Box

Lees-Miller

Snowdon

et al. 2013

16th International IEEE Conference on Intelligent Transportation Systems (ITSC 2013)

Self Cite

View full text Add to dashboard Cite

Abstract-An experiment was conducted using the InnovITS proving ground in Nuneaton. Thirty cars with volunteer drivers were asked to drive around a tight closed road circuit causing them to pass repeatedly through a cross-roads junction from all directions. The junction was signalized. In different testruns of the experiment the traffic lights were controlled by either an automated fixed-time system or by a human using remote control. All vehicles in the test were instrumented using GPS and bluetooth. Video footage from two cameras was also recorded.The goal of the experiment was to collect data on the performance of human junction controllers. This was motivated by earlier work indicated that human controllers could perform well at this task in a simulated 'computer game' environment.In particular this paper examines some of the issues that arise when trying to simulate an urban road junction in this manner. For example results are presented indicating differences in network performance depending on whether the drivers were instructed to follow a fixed route or a random route of their choice. Thus providing some guidance for maximizing the fidelity of this type of simulation in the future.The paper also presents a detailed analysis of the sensor data and video footage to measure the performance of the junction under the different modes of control.

show abstract

Section: ) Control Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Lessons from proving ground experiments to investigate junction control

Box

Lees-Miller

Snowdon

et al. 2013

16th International IEEE Conference on Intelligent Transportation Systems (ITSC 2013)

Self Cite

View full text Add to dashboard Cite

show abstract

“…In this paper we present an adaptation to the neural network based junction control system described in Box and Waterson (2012). This adaptation enables the controller to be trained under simulation by temporal difference reinforcement learning.…”

Section: Context and Motivationmentioning

confidence: 99%

An automated signalized junction controller that learns strategies by temporal difference reinforcement learning

Box

Waterson

2013

Engineering Applications of Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

This paper shows how temporal difference learning can be used to build a signalized junction controller that will learn its own strategies though experience. Simulation tests detailed here show that the learned strategies can have high performance. This work builds upon previous work where a neural network based junction controller that can learn strategies from a human expert was developed . In the simulations presented, vehicles are assumed to be broadcasting their position over WiFi giving the junction controller rich information. The vehicle's position data are pre-processed to describe a simplified state. The state-space is classified into regions associated with junction control decisions using a neural network. This classification is the strategy and is parametrized by the weights of the neural network. The weights can be learned either through supervised learning with a human trainer or reinforcement learning by temporal difference (TD). Tests on a model of an isolated T junction show an average delay of 14.12 s and 14.36 s respectively for the human trained and TD trained networks. Tests on a model of a pair of closely spaced junctions show 17.44 s and 20.82 s respectively. Both methods of training produced strategies that were approximately equivalent in their equitable treatment of vehicles, defined here as the variance over the journey time distributions.

show abstract

“…The authors of [11] make explicit reference to the idea of neurons. The neural network used to model the traffic control system is trained by a live expert-a traffic engineer.…”

Section: Traffic Controlmentioning

confidence: 99%

Neural Networks in Transportation Research – Recent Applications

Pamuła¹

2017

Transport Problems

View full text Add to dashboard Cite

Summary. Neural networks' (NNs) capability of mapping the nonlinear functions of variables describing the behaviour of objects and the simplicity of designing their configuration favours their applications in transport. This paper presents representative examples in the scope of prediction of road traffic parameters, road traffic control, measurement of road traffic parameters, driver behaviour and autonomous vehicles, and transport policy and economics. The features of the solutions are examined. The review shows that feedforward multilayer neural networks are the most often utilised configurations in transportation research. No systematic approach is reported on the optimisation of the NN configurations to achieve a set level of performance in solving modelling tasks. ZASTOSOWANIE SIECI NEURONOWYCH W BADANIACH W TRANSPORCIE -OSTATNIE OPRACOWANIAStreszczenie. Zdolność sieci neuronowej do odwzorowania nieliniowych zależności między zmiennymi, które opisują zachowanie obiektów, oraz łatwość opracowania efektywnej konfiguracji sprzyjają zastosowaniom ich w transporcie. W artykule przedstawiono reprezentatywne przykłady z zakresu: predykcji parametrów ruchu drogowego, sterowania ruchem drogowym, pomiarów parametrów ruchu, zachowania kierowców i prowadzenia autonomicznych pojazdów, ekonomii i polityki transportowej, oraz omówiono własności proponowanych rozwiązań. Przegląd wskazuje, że najczęściej wybieranymi sieciami neuronowymi są sieci jednokierunkowe wielowarstwowe. Brak jest systematycznego podejścia do optymalizacji konfiguracji sieci w celu osiągnięcia zadanego poziomu dokładności w zadaniach modelowania.

show abstract

An automated signalized junction controller that learns strategies from a human expert

Cited by 13 publications

References 5 publications

Lessons from proving ground experiments to investigate junction control

Lessons from proving ground experiments to investigate junction control

An automated signalized junction controller that learns strategies by temporal difference reinforcement learning

Neural Networks in Transportation Research – Recent Applications

Contact Info

Product

Resources

About