Vehicle platooning can be formulated as an optimal control problem and many solving paradigms, such as Pontryagin's maximum principle-based and dynamical programming methods, have been recently developed. However, these methods usually rely on solving a group of necessary conditions or Hamilton-Jacobi-Bellman (HJB) partial differential equations, which is hard to calculate. Besides, due to the heterogeneous dynamics of different vehicles in a mixed and complex platoon which comprises of not only connected autonomous vehicles (CAVs), but also human-driven vehicles (HDVs), it is also challenging to coordinate the behaviors of different vehicles in an unified control framework. Here we provide a Neural Mixed Platoon Control (NMPC) framework, a novel control design for mixed vehicle platooning based on a neural ordinary differential equation (NODE). We first formulate an optimal control model that incorporates the heterogeneous dynamics of a leading CAV and several following HDVs. We use a neural network to parameterize a state-feedback controller and join the neural controller and the mixed platooning dynamics into the NODE solver to create a closed-loop and learnable controlled system. The resulting system can learn optimal control inputs driving the mixed platoon to evolve from a given beginning condition to the target state within a finite duration in an unsupervised manner. Finally, simulation results validate our suggested method's usefulness in terms of space headway and velocity tracking.
Neural Mixed Platoon Controller Design
28.10.2022
589025 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch