Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quad-LCD: Layered Control Decomposition Enables Actuator-Feasible Quadrotor Trajectory Planning

Anusha Srikanthan, Hanli Zhang, Spencer Folk, Vijay Kumar, Nikolai Matni

TL;DR

This work tackles quadrotor trajectory planning under motor saturation, where aggressive maneuvers cause uncontrolled drift and crashes. It introduces Quad-LCD, a data-driven planner that uses a layered control decomposition to produce actuator-feasible reference trajectories while a learned tracking cost $g^{ctrl}$ biases the plan toward feasible maneuvers under a fixed nonlinear controller. Trajectories are represented as piecewise polynomials with a minimum-snap objective, augmented by the learned cost, enabling safer planning without changing the controller itself. In simulation, the approach reduces crash rates by about $49\%$ compared with baselines, and hardware experiments on Crazyflie demonstrate zero-shot sim-to-real feasibility, supported by an open-source ROS Python interface for deployment.

Abstract

In this work, we specialize contributions from prior work on data-driven trajectory generation for a quadrotor system with motor saturation constraints. When motors saturate in quadrotor systems, there is an ``uncontrolled drift" of the vehicle that results in a crash. To tackle saturation, we apply a control decomposition and learn a tracking penalty from simulation data consisting of low, medium and high-cost reference trajectories. Our approach reduces crash rates by around $49\%$ compared to baselines on aggressive maneuvers in simulation. On the Crazyflie hardware platform, we demonstrate feasibility through experiments that lead to successful flights. Motivated by the growing interest in data-driven methods to quadrotor planning, we provide open-source lightweight code with an easy-to-use abstraction of hardware platforms.

Quad-LCD: Layered Control Decomposition Enables Actuator-Feasible Quadrotor Trajectory Planning

TL;DR

This work tackles quadrotor trajectory planning under motor saturation, where aggressive maneuvers cause uncontrolled drift and crashes. It introduces Quad-LCD, a data-driven planner that uses a layered control decomposition to produce actuator-feasible reference trajectories while a learned tracking cost biases the plan toward feasible maneuvers under a fixed nonlinear controller. Trajectories are represented as piecewise polynomials with a minimum-snap objective, augmented by the learned cost, enabling safer planning without changing the controller itself. In simulation, the approach reduces crash rates by about compared with baselines, and hardware experiments on Crazyflie demonstrate zero-shot sim-to-real feasibility, supported by an open-source ROS Python interface for deployment.

Abstract

In this work, we specialize contributions from prior work on data-driven trajectory generation for a quadrotor system with motor saturation constraints. When motors saturate in quadrotor systems, there is an ``uncontrolled drift" of the vehicle that results in a crash. To tackle saturation, we apply a control decomposition and learn a tracking penalty from simulation data consisting of low, medium and high-cost reference trajectories. Our approach reduces crash rates by around compared to baselines on aggressive maneuvers in simulation. On the Crazyflie hardware platform, we demonstrate feasibility through experiments that lead to successful flights. Motivated by the growing interest in data-driven methods to quadrotor planning, we provide open-source lightweight code with an easy-to-use abstraction of hardware platforms.
Paper Structure (8 sections, 4 equations, 4 figures)

This paper contains 8 sections, 4 equations, 4 figures.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Optimizing for Actuator-Feasible Polynomial Trajectories
  4. Learning tracking cost from simulation data
  5. Simulation and Experiments
  6. Experimental Setup for Hardware
  7. Conclusion
  8. Acknowledgment

Figures (4)

  • Figure 1: We show the flight path of a Crazyflie 2.0 resulting in a crash due to motor saturation from aggressive flight on the top and successful flight below. The reference paths were designed using a baseline planner mellinger2011minimum and our approach, respectively.
  • Figure 2: A visualization of low, medium and high-cost trajectories simulated on RotorPy.
  • Figure 3: We show a visualization of trajectories simulated on RotorPy for the standard Crazyflie platform where dotted and solid lines are reference and controller executed trajectories, respectively. On the left is a 3D plot showing the deviation of controller executed trajectories and reference for our approach and a baseline. On the right, we plot the $x, y, z$ and $\psi$ curves with waypoints. The deviations in $x$ and $y$ for the baseline planner is high due to large swings from controller saturation while Ours-GC plans references that are tracked more accurately.
  • Figure 4: We report crash rates evaluated on $100$ waypoint-following tasks and average segment speed of $2 (m/s)$. The asterisk ($*$) denotes that motor noise was turned off for training and evaluation on the RL policy.