In this project, we present a few demos of DeeP-LCC for Cooperative Control of Connected and Autonomous Vehicles (CAVs) in mixed traffic. The paper can be found here.
See our experimental validation on youtube.
See Distributed DeeP-LCC for a distributed version of DeeP-LCC, which is applicable to large-scale mixed traffic control.
DeeP-LCC is a data-driven predictive control strategy for CAVs in mixed traffic, where human-driven vehicles (HDVs) also exist. Our strategy aims to deal with unknown nonlinear car-following behaviors of HDVs.
Insead of assuming a parametric car-following model, DeeP-LCC directly relies on measurable driving data to achieve safe and optimal control for CAVs. It is adapted from the standard Data-EnablEd Predictive Control (DeePC) method considering the characteristics of mixed traffic. DeeP-LCC is implemented in a receding horizon manner, in which input/output constraints are incorporated to achieve collision-free guarantees.
Related projects:
DeeP-LCC collects three types of trajectory data from mixed traffic:
- Control input : acceleration signal of the CAVs;
- Traffic output : velocity error of all the following vehicles (including HDVs and CAVs), and spacing error of the CAVs;
- External input : velocity error of the had vehicle.
The following optimization problem is converted to quadratic programming for problem solving, and is implemented in a receding horizon manner.
We carry out experimental validations on a miniature traffic platform.
The complete video can be found on youtube.
Case 1: all the vehicles are HDVs.
Case 2: Vehicle no.2 utilizes DeeP-LCC.
Case 3: Vehicles no.2 and no.4 utilize DeeP-LCC.
To contact us about DeeP-LCC, email either Jiawei Wang or Yang Zheng.