APACRace

Code for Adaptive Planning and Control with Time-Varying Tire Models for Autonomous Racing Using Extreme Learning Machine (Website) (Paper) NOTE: Well formatted and cleaned version of the code will be available soon

Overview

Installation

We recommend creating a new conda environment:

conda create --name apacrace python=3.6
conda activate apacrace

Then install APACRace:

git clone https://github.com/dvij542/APACRace-Code.git
cd APACRace-Code/
pip install -e .

How to run online learning based controller (APACRace)

The following steps are explained for the 1:43 scale autonomous racing platform at ETH Zurich. We also provide code for the 1:10 scale F1TENTH racing platform at University of Pennsylvania.

1. (Optional, generated raceline is in repo) Compute the racing line for the track we want to race on.

   cd apacrace/raceline
   python generate_raceline_ethz.py
   

2. (Optional, generated raceline library is in repo) Render speeds along the raceline for different tire friction limits by running :-

   python render_speeds.py
   

3. Run APACRace (MLP predictor for tire forces learned online) controller with the generated offline library of racelines as reference as follows:-

   cd mpc
   python run_nmpc_online_semi_mlp.py
   

4. (optional) Run the following optionally for comparison:- a. Full dynamics learning with MLP: run_nmpc_online_mlp.py b. Using Gaussian Process (GP) for model correction: run_nmpc_online_gp.py c. Dynamic model with known parameters (Oracle): run_nmpc_true.py d. E-kinematic model with no learning: run_nmpc_ekin.py

How to run offline GP-based controller (BayesRace)

The following steps are explained for the 1:43 scale autonomous racing platform at ETH Zurich. We also provide code for the 1:10 scale F1TENTH racing platform at University of Pennsylvania.

1. Compute the racing line for the track we want to race on.

   cd apacrace/raceline
   python generate_raceline_ethz.py
   

2. Run a pure pursuit controller on a different track to log sensor measurements and state estimates. This data resemble true system dynamics.

   cd apacrace/pp
   python run_pp_orca.py
   

3. Given a trajectory of states from Step 2, generate an equivalent trajectory using a simpler and easy-to-tune e-kinematic model.

   cd apacrace/data
   python simulate_kinematic_orca.py
   

4. Train Gaussian process models to predict mismatch between true system dynamics from Step 2 and e-kinemtic model from Step 3.

   cd apacrace/gp
   python train_model_orca.py
   

5. Validate the trained models on the track we want to race.

   cd apacrace/gp
   python plot_uncertainty_orca.py
   

6. Run MPC with and without GP correction. By default boundary constraints are turned off for faster execution.

   cd apacrace/mpc
   python run_nmpc_orca_gp.py
   

   MPC without GP correction is shown on the left, and with GP correction on the right.

7. Benchmark the performance in Step 6 against MPC with true model dynamics.

   cd apacrace/mpc
   python run_nmpc_orca_true.py
   

8. Finally, update the GP models using data collected in Step 6 that is specific to the racing track and re-run MPC with the updated GP models.

   cd apacrace/gp
   python update_model_orca.py
   

   cd apacrace/mpc
   python run_nmpc_orca_gp_updated.py