Rethinking Reference Trajectories in Agile Drone Racing: A Unified Reference-Free Model-Based Controller via MPPI
Code (coming soon) arXivAbstract
While model-based controllers have demonstrated remarkable performance in autonomous drone racing, their performance is often constrained by the reliance on pre-computed reference trajectories. Conventional approaches, such as trajectory tracking, demand a dynamically feasible, full-state reference, whereas contouring control relaxes this requirement to a geometric path but still necessitates a reference. Recent advancements in reinforcement learning (RL) have revealed that many model-based controllers optimize surrogate objectives, such as trajectory tracking, rather than the primary racing goal of directly maximizing progress through gates. Inspired by these findings, this work introduces a reference-free method for time-optimal racing by incorporating this gate progress objective, derived from RL reward shaping, directly into the Model Predictive Path Integral (MPPI) formulation. The sampling-based nature of MPPI makes it uniquely capable of optimizing the discontinuous and non-differentiable objective in real-time. We also establish a unified framework that leverages MPPI to systematically and fairly compare three distinct objective functions with a consistent dynamics model and parameter set: classical trajectory tracking, contouring control, and the proposed gate progress objective. We compare the performance of these three objectives when solved via both MPPI and a traditional gradient-based solver. Our results demonstrate that the proposed reference-free approach achieves competitive racing performance, rivaling or exceeding reference-based methods.
Framework Overview
A conceptual illustration of the unified MPPI framework. (a) The core sampling and weighting stage of MPPI. (b) A comparison of the proposed reference-free gate progress objective with two conventional reference-based objectives: Trajectory Tracking and Contouring Control.
The left panel shows states within the predictive horizon being assigned different target gate indices (using different color) to achieve rapid progress along the track. To ensure the stable switching, we propose the temporally consistent strategy shown on the right, where an index switch only occurs for states that pass through the gate between two consecutive control loops. This logic, combined with the gate progress objective enables reference-free racing.
Video Presentation
Results
Simulation Results
Flight trajectories generated by our unified MPPI framework on three tracks of increasing difficulty: Circle (a-c), Figure-8 (d-f), and Split-S (g-i). All the flight trajectories are tested under identical system dynamics and controller settings.
Comparison across different objectives. 'Opt.' refers to the gradient-based OCP solver, while 'Ours' refers to our MPPI-based implementation.
Real-World Experiments
Real-world flight validation of the three control objectives executed within our unified MPPI framework. The panels depict the quadrotor's performance using: (left) Trajectory Tracking, (middle) Contouring Control, and (right) the proposed reference-free Gate Progress objective. All three strategies were deployed using identical dynamics parameters and a consistent control input structure.
Videos
BibTeX
@misc{zhao2025rethinkingreferencetrajectoriesagile,
title={Rethinking Reference Trajectories in Agile Drone Racing: A Unified Reference-Free Model-Based Controller via MPPI},
author={Fangguo Zhao and Xin Guan and Shuo Li},
year={2025},
eprint={2509.14726},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2509.14726},
}