APACE: Agile and Perception-Aware Trajectory Generation for Quadrotor Flights

TL;DR

APACE, an Agile and Perception-Aware trajeCtory gEneration framework for quadrotors aggressive flight, that takes into account feature matchability during trajectory planning and proposes a differentiable and accurate visibility model that allows decomposition of the trajectory planning problem for efficient optimization resolution.

Abstract

Various perception-aware planning approaches have attempted to enhance the state estimation accuracy during maneuvers, while the feature matchability among frames, a crucial factor influencing estimation accuracy, has often been overlooked. In this paper, we present APACE, an Agile and Perception-Aware trajeCtory gEneration framework for quadrotors aggressive flight, that takes into account feature matchability during trajectory planning. We seek to generate a perception-aware trajectory that reduces the error of visual-based estimator while satisfying the constraints on smoothness, safety, agility and the quadrotor dynamics. The perception objective is achieved by maximizing the number of covisible features while ensuring small enough parallax angles. Additionally, we propose a differentiable and accurate visibility model that allows decomposition of the trajectory planning problem for efficient optimization resolution. Through validations conducted in both a photorealistic simulator and real-world experiments, we demonstrate that the trajectories generated by our method significantly improve state estimation accuracy, with root mean square error (RMSE) reduced by up to an order of magnitude. The source code will be released to benefit the community.

Figures (7)

• Figure 1: A real-world aggressive flight experiment conducted in a challenging darkness environment. The LEDs on the quadrotor are only for photography purpose and disabled during the experiments. The bottom row shows snapshots of quadrotor's first-person view at the corresponding circled positions, where the red dots represent the matched features.
• Figure 2: An overview of the proposed trajectory generation method.
• Figure 3: Visualization of the sigmoid functions (\ref{['eq:v1']}) and (\ref{['eq:v2']}) with $\alpha_v = 60^{\circ}$ and different $k$ values where the black dashed lines indicate the corresponding step functions to be approximated. The brown dotted curve in the left plot illustrates the constant weight $v_1^\prime(\theta_1)$ used in (\ref{['eq:f_para']}).
• Figure 4: Illustration of the proposed visibility model under FoV setting $(\alpha_h, \alpha_v) = (90^\circ, 60^\circ)$. The violet shape in (a-c) represents visible regions. (d) visualizes the proposed visibility model by dense sampling. Points with visibility model take values close to zero are transparent and colored blue, and values close to one are opaque and colored pink.
• Figure 5: The setup in the photorealistic simulator Airsim is shown in (a). The trajectories generated by PAGzhou2019robust, PAWMurali2019 with top velocity $2.0 m\slash s$ and our planner with top velocity $4.0 m\slash s$ along with the visual odometries (VO) are illustrated from side and top views in (b) and (c) respectively. The yaw directions are visualized as arrows at interval of 0.2s.