Model Predictive Control of Autonomous Vehicles Incorporating Trajectories Predicted by Deep Learning (DL-MPC)

Authors: Zengjie Zhang (z.zhang3@tue.nl), Ni Dang (ni.dang@tum.de)

A demonstration of controlling an ego autonomous vehicle incorporating the target vehicle trajectory predicted by a recurrent neural network. For more details, refer to our ArXiv article at https://arxiv.org/pdf/2310.02843.

Quick Guide

Scenario of demonstration

This demonstration considers a highway scenario where an agent-driving ego vehicle moves along the central lane while avoiding collisions with a human-driving target vehicle cutting in from the slow lane, as shown in Fig. 1. A recursive neural network with long short-term memory (LSTM) units is trained to predict the trajectory of the target vehicle according to its historical trajectories. Based on the predicted trajectory, a model predictive control (MPC) is used to allow safe maneuvering for the ego. Fig. 1 showcases the resulting trajectories of both vehicles. Fig. 1 is an example of trajectory prediction.

Note: for the current version, the target vehicle is hardcoded using an 'S' shape spline.

Fig 1. the highway driving scenario.

Fig 2. prediction example of the target vehicle trajectory.

Environment Requirements

• Windows: tested on 11 but compatible in general
• CUDA (Optional): tested on 12.2
• MATLAB: tested on R2022b but compatible in general
  • Deep Learning Toolbox: version 14.5
  • Parallel Computing Toolbox (Optional): version 7.7

File Structure

The files are organized as the following structure:

README.md                   % This read-me file
LICENSE                     % The license file
main.m                      % The main code script
config/
    param.m                 % The script to render parameters
libs/
    controller              % Functions related to control
    collision               % Functions related to collision detection
    draw                    % Functions related to visualization
    training                % Functions related to predictor training

Quick Run

• Run the main script main.m;
• The rnn predictor and its training information will be saved in policy/;
• The control results will be saved in data/;
• All figures will be saved in figures/.