4 présentations

• 
  15h30 - 15h55

  Model-predictive control of autonomous mobility-on-demand systems

  • Ramon Iglesias, prés., Stanford University
  • Marco Pavone, Stanford University
  • Matthew Tsao, Stanford University
  • Federico Rossi, NASA Jet Propulsion Lab

  In this talk, we present a model-predictive control framework for Autonomous Mobility-on-Demand (AMoD) systems. The framework consists of a forecasting generative model and a stochastic optimization subproblem. We show via simulation that this approach vastly outperforms state-of-the-art fleet-level control algorithms and is more robust with respect to uncertain demand.
• 
  15h55 - 16h20

  On the interaction between autonomous mobility-on-demand and the urban environment

  • Mauro Salazar, prés., Stanford University
  • Kiril Solovey, Stanford University
  • Maximilian Schiffer, TU München
  • Marco Pavone, Stanford University

  This talk presents models and coordination policies for Autonomous Mobility-on-Demand (AMoD), wherein a fleet of self-driving vehicles provides on-demand mobility, potentially jointly with public transit. I will focus on the application of optimization methods to devise routing strategies for AMoD systems and assess their potential benefits.
• 
  16h20 - 16h45

  Control of autonomous electric fleets for ridehail systems

  • Nicholas Kullman, prés., CIRRELT
  • Martin Cousineau, HEC
  • Justin C. Goodson, Saint Louis University
  • Jorge E. Mendoza, CIRRELT

  We consider a ridehail company operating a fleet of autonomous electric vehicles. The operator assigns vehicles to new requests and repositions/recharges vehicles in anticipation of future requests. We model the problem as an MDP, contrast solutions from deep reinforcement learning and approximate dynamic programming, and offer a dual bound.

  Keywords: Autonomous vehicles, Markov decision process, deep reinforcement learning
• 
  16h45 - 17h10

  Multi-period electric vehicle routing and charging scheduling problems

  • Laura C. Echeverri, prés., LIFAT
  • Aurélien Froger, Inria
  • Jorge E. Mendoza, CIRRELT
  • Emmanuel Néron, LIFAT

  We consider a fleet of electric vehicles (EVs) that must serve customers over several days. EVs are charged at the depot, subject to the charging infrastructure constraints. We consider the effect of operational conditions on EV battery aging. We propose MILP formulations and a matheuristic approach to solve this problem. / Electric vehicle, battery degradation, mixed integer linear programming