Why Change Your Controller When You Can Change Your Planner: Drag-Aware Trajectory Generation for Quadrotor Systems

TL;DR

The paper tackles quadrotor payload delivery under aerodynamic wrenches by proposing drag-aware trajectory generation that relaxes the original optimal control problem and incorporates a learned tracking-penalty regularizer $g_{ ho,\pi}^{track}$ into planning. A layered, data-driven decomposition yields a planner that explicitly accounts for the SE(3) controller's ability to track reference trajectories, avoiding the pitfalls of purely reactive controller redesign. Key contributions include formulating the drag-aware planning problem, learning the controller tracking cost via policy evaluation using Monte Carlo data, and validating the approach in both RotorPy simulations and Crazyflie hardware with substantial improvements in tracking accuracy (up to 83% reduction in position error) and safer aggressive maneuvers. The results demonstrate that proactive planning to mitigate drag effects can significantly enhance performance and safety in aerial delivery tasks, with open-source code and plans for faster convex approximations in future work.

Abstract

Motivated by the increasing use of quadrotors for payload delivery, we consider a joint trajectory generation and feedback control design problem for a quadrotor experiencing aerodynamic wrenches. Unmodeled aerodynamic drag forces from carried payloads can lead to catastrophic outcomes. Prior work model aerodynamic effects as residual dynamics or external disturbances in the control problem leading to a reactive policy that could be catastrophic. Moreover, redesigning controllers and tuning control gains on hardware platforms is a laborious effort. In this paper, we argue that adapting the trajectory generation component keeping the controller fixed can improve trajectory tracking for quadrotor systems experiencing drag forces. To achieve this, we formulate a drag-aware planning problem by applying a suitable relaxation to an optimal quadrotor control problem, introducing a tracking cost function which measures the ability of a controller to follow a reference trajectory. This tracking cost function acts as a regularizer in trajectory generation and is learned from data obtained from simulation. Our experiments in both simulation and on the Crazyflie hardware platform show that changing the planner reduces tracking error by as much as 83%. Evaluation on hardware demonstrates that our planned path, as opposed to a baseline, avoids controller saturation and catastrophic outcomes during aggressive maneuvers.

Figures (3)

• Figure 1: We show a visualization of trajectories for $\Bar{\rho} = 0$ simulated on RotorPy where dotted and solid lines are reference and controller executed trajectories, respectively. On the left are 3D plots showing the deviation of controller executed trajectories and reference for our approach and baselines. On the right, we plot the $x, y, z$ and $\psi$ curves with waypoints to show the deviation in $x$ for baseline trajectories due to controller saturation while the drag-aware plan is tracked more accurately.
• Figure 2: On the left, we plot cumumlative tracking error vs time for our approach and baselines for the trajectories visualized in Fig. \ref{['fig:3d-plots']}. On the right, we plot the relative cost ratio of our approach over "minsnap" baseline. The orange line represents the median ratio, separating the lower quartile from the upper quartile. The whiskers are the extreme values. We note that the tracking costs are obtained from the dynamics simulation (true cost) and not the network predictions.
• Figure 3: We visualize snapshots of demonstrating trajectories planned by "minsnap" as shown in a) and our approach as shown in b) on Crazyflie 2.0 and observe that the maneuver planned by the baseline results in controller saturation and crashing the quadrotor while our method planned a feasible path for the controller.