This repository implements a competitive two-player vehicle racing and a solver for nonlinear, constrained bilevel optimization.
We set up two-player racing as an equilibrium problem. We define cost and constraints to describe simplified aerodynamic drag, aerodynamic drafting, position-dependent collision-avoidance, and the road. We use our model to explore how different solution concepts affect competitiveness.
This repository supplements our corresponding conference paper at ICRA 2025.
Simple usage example is available in examples/racing_example.jl
using EpecRacing
probs = setup()
sim_results = solve_simulation(probs, time_steps; x0, road, mode):The solution concepts are enumerated using the mode ID, defined as:
| Mode ID | Player 1 | Player 2 |
|---|---|---|
| 1 | SP | SP |
| 2 | SP | NE |
| 3 | NE | NE |
| 4 | SP | Leader |
| 5 | NE | Leader |
| 6 | Leader | Leader |
| 7 | SP | Follower |
| 8 | NE | Follower |
| 9 | Leader | Follower |
| 9 | Follower | Follower |
Simulation results can be visualized using:
animate(probs.params, sim_results; mode, road)Utilities to set up and solve experiments (batch run of simulations) are also available. There is an example on how to do this in examples/experiment_example.jl:
probs, x0s, roads, params = create_experiment(sample_size)
results, elapsed = solve_experiment(probs, x0s, roads, time_steps, mode)Distributed under the MIT License. See LICENSE for more information.
@misc{cinar2024doesbileveloptimizationresult,
title={Does bilevel optimization result in more competitive racing behavior?},
author={Andrew Cinar and Forrest Laine},
year={2024},
eprint={2402.09548},
archivePrefix={arXiv},
primaryClass={cs.GT},
url={https://arxiv.org/abs/2402.09548},
}