Section: New Results

Distributed Agreement and String Control in Intelligent Vehicular Networks (IVNs)

Participant : Gérard Le Lann.

IVNs are composed of automated (autonomous and communicating) vehicles, ranging from pre-planned platoons to ad hoc vehicular networks (VANETs). Agreement problems in the presence of concurrency and failures are not well investigated yet in IVNs. We have examined a specific class of such problems, those arising in string formations. Regarding string membership (vehicles leaving or joining a string), with few exceptions, safety issues have been addressed so far assuming that (1) no more than 1 insertion operation would be performed at any given time or, (2) every vehicle decides unilaterally, i.e. undertakes a maneuver after having activated some signal, leaving to surrounding vehicles the responsibility of inferring intended maneuvers. Assumption (1) is not realistic. There are numerous risk-prone scenarios where a posteriori reactive approaches (assumption (2)) may fail. Therefore the need for investigating proactive approaches, where vehicles (1) are made aware of intended impending maneuvers, (2) agree on which maneuvers can be safely undertaken, prior to performing physical maneuvers. It follows that a solution to numerous string control problems consists of a pair ($A$,$\Phi$), where $A$ stands for a distributed agreement algorithm which achieves global coordination in the presence of failures and concurrency, and $\Phi$ stands for control laws drawn from control theory and robotics. Algorithm $A$ is run prior to local activations of $\Phi$.

Our work is based on the cohort construct, which serves to formalize the concept of strings. Velocity Agreement is the generic problem selected. At any time, some number of string/cohort members may propose each a new velocity value. In fine, every vehicle computes a unique new velocity $V$. Proposed values are propagated via neighbor-to-neighbor (N2N) radio communications. We have devised a solution called the VAgree algorithm. In the presence of up to $f$ failures (on-board systems, N2N message losses), the following properties shall hold:

• Validity: Decision value $V$ = $\Psi$(proposed values).

• Agreement: No two members decide differently.

• Time-Bounded Termination: VAgree terminates at most $\theta$ time units.

• Synchronicity: Times at which $V$ is posted to on-board systems are comprised within a small time interval $ϵ$. Distance traveled during $ϵ$ by the member earliest to post $V$ until the latest member does so is an order of magnitude smaller than vehicle sizes.

The VAgree algorithm is presented in a paper which is under submission.