Section: New Results

Transportation networks and vehicular systems

Traffic estimation and prediction

Participants : C. Canudas de Wit [Contact person] , A. Kibangou, L. Leon Ojeda, F. Morbidi.

Reconstructing densities in portions of the road links not equipped with sensors constitutes an important task in traffic estimation, forecasting, and control problems. Among many other approaches, model-based observers is one popular technique to build this information. They can also be understood as virtual sensors deployed inside of the cells not equipped with true sensors. They are used to better track, in real-time, density variations with a fine degree of granularity in the space, as the virtual cells can be selected as small as desired.

In [43] , a graph constrained-CTM observer was introduced. It allows reconstructing rather accurately the internal states (densities) of a road portion not equipped with sensors. This strategy for real-time density estimation was applied on Grenoble South Ring. In [27] , this observer has been associated with an adaptive Kalman filtering approach for traffic prediction in terms of travel time. The adaptive Kalman filtering approach was also been used for predicting input flows in [26] .

Traffic control

Participants : C. Canudas de Wit [Contact person] , D. Pisarski.

This work has been conducted in two parallel directions, combining steady state analysis and design of an optimal ramp metering controller.

The first direction was to extend the preliminary results presented in the papers [10] and [46] . The goal was to implement the idea of optimal steady state balancing. A relevant software was built up and tested on the model representing the south ring of Grenoble. The results were published in [28] . A comprehensive study of steady state balancing was submitted as a journal paper, under review.

The second direction was to develop a distributed optimal ramp metering controller. This study is motivated by two following facts. The first one is to decentralize and parallelize computation for optimal freeway traffic control problem, and thus to reduce computational complexity. The second one is to reduce the lengths of the communication channels, in order to eliminate the probability of information delay or packet loss. The proposed new control objective provides a uniformly distributed (or balanced) vehicle density such that the usage of freeway (measured by the Total Travel Distance and Total Input Volume) is maximized. This optimal balancing objective is reached by taking a proper state feedback control structure and optimizing the set of gains. Here we imposed distributed condition for both, the feedback structure and the optimization process. We have focused the efforts to design the controller network architecture that is based on the common elements (ramp meter controllers), executing the same computational procedures and applying the control signals based on the same state-feedback structure. This meets a spirit of `plug and play' (PnP), and is beneficial for both, architecture assembling and component replacement (in case of failure). In order to define the functionality for each of this PnP controller, the analysis on both system controllability and conditions for optimality were carried on. The preliminary work let us to determine the what type of information and upon which communication topology it is required to be sent in order to solve the posed optimization problem. Firstly, the feedback controls for each of the controllers require state information of the section that is controllable for it. In general, each of the controllers demands the state for its closest surrounding sections (downstream and upstream). Secondly, each of the controllers communicates with its closest active neighbors to exchange the information of optimal solution, namely optimal boundary flow or optimal control. We also observed that in any system mode there might be only one inactive controller (the controller that does not have any controllable section) surrounded by two active ones, and thus the maximum required information comes from the two closest neighbors for each of the directions. We noticed also that inactive controllers may serve to convey the information for the active ones, so the communication can be based on a path (or linear) graph. Part of this research was realized in UC Berkeley during the visit of Dominik Pisarski in PATH laboratory.

Traffic control is based on models of traffic, usually the so-called CTM – Cell transmission Model. Some work in the team aims at developing different models, more suitable for control. One such model is based on cells of variable length, as an alternative way to describe the congestion position. This model, proposed in [42] , has been refined in the master thesis of Giulio Bontadini, taking into account mass conservation laws.

Vehicle control for disabled people

Participants : C. Canudas de Wit [Contact person] , V. Ciarla, J. Dumon, F. Quaine [UJF] , V. Cahouet [UJF] .

Disabled people face the effort to turn the steering wheel while driving their vehicle. This study, funded by the VolHand project, focuses on the aspect of the assistance during driving maneuver at low speeds (for instance, parking). On common vehicles for healthy people, the system that improves the driver's steering feel in these situations is the power-steering stage, which is mounted at the basis of the steering column and is based on hydraulic technology; the new generation uses an electric motor instead of the hydraulic pump, with more advantages in terms of fuel consumption, better road-feel feedback to the driver and better return-to-center performances of the steering wheel. This work has developed a general methodology to adapt the current technology for disabled people, by introducing additional blocks that can be implemented via software without altering the hardware of the vehicle. In this way, it can be easily exported without additional costs in terms of design and technology for the industrial partner. The methodology has been studied theoretically, joining control aspects with bio-mechanical ones. Moreover, the theoretical study has been tested in laboratory on the hardware-in-the loop setup, using the experimental platform NeCSCar (see Section 5.2 ). First, a real steering wheel has been linked to a real-time PC-unit and to an electrical motor. A graphical user interface has been implemented to facilitate the access to the software. Then, the last part of the study has been the experimental validation with a tele-operated real vehicle. The vehicle provided the measure of the friction torque to the PC-unit, simulating a real driving situation.

This work is described in [41] and in the Ph.D. thesis [11] .

Control of communicating vehicles in urban environment

Participants : C. Canudas de Wit [Contact person] , G. de Nunzio.

For a given vehicle there are different ways to travel on a given distance in a given time, associated to different levels of energy consumption; therefore, it is always possible to find an energy-optimal trajectory. Advising the driver via a suitable interface can reduce the energy consumed during the travel, and thus improve the energy efficiency: this is the principle of eco-driving. In urban areas, the optimal trajectory of the vehicle depends on interactions with other vehicles, also on passive signs (panels, priorities, etc.) and active signs (traffic lights); in each case, constraints are imposed on the command (vehicle speed). From the infrastructure perspective, traffic control in urban areas consists in determining the state of traffic signals in order to solve an optimization problem, for example minimizing average travel time of vehicles in the road network. If all vehicles could communicate with one another and with the active infrastructure (traffic lights), we could imagine benefits for each of the two problems which can be considered as a whole: on the one hand, from the vehicles' point of view, more information is available that can be integrated into the online optimization problem; on the other hand, there are new measures and new commands available to control traffic. Indeed, the estimation of the traffic is no longer necessary, as the position and speed of approaching vehicles is known and shared. More importantly, the traffic manager can send instructions to the vehicles. The aim of the research is to evaluate the potential in terms of energy saving and traffic improvement made possible by communicating vehicles. This work is the topic of the Ph.D. thesis of Giovanni De Nunzio, a CIFRE thesis with IFPEN. The paper [21] considers the scenario where vehicle and infrastructure (traffic lights) can communicate, and describes a suitable optimization algorithm that can be run on-board the vehicle so to optimize its energy consumption by avoiding stops and abrupt changes of speed at traffic lights, thanks to the information on upcoming traffic lights on the same road.