Eva-Tracker: ESDF-update-free, Visibility-aware Planning with Target Reacquisition for Robust Aerial Tracking

TL;DR

Eva-Tracker tackles the problem of maintaining target visibility for aerial tracking under cluttered 3D environments with limited onboard compute. It replaces real-time ESDF updates by a precomputed Field of View ESDF (FoV-ESDF) and couples it with a recovery-capable, visibility-aware trajectory framework that includes target trajectory prediction and an efficient initial path generation. The approach unifies occlusion avoidance, observation-distance control, and viewing-angle alignment into differentiable FoV-ESDF-based objectives, and supports fast reacquisition when the target is temporarily lost. Extensive simulations and real-world flights demonstrate improved robustness and lower computational cost compared with state-of-the-art ESDF-based methods, enabling practical real-time operation on onboard hardware.

Abstract

The Euclidean Signed Distance Field (ESDF) is widely used in visibility evaluation to prevent occlusions and collisions during tracking. However, frequent ESDF updates introduce considerable computational overhead. To address this issue, we propose Eva-Tracker, a visibility-aware trajectory planning framework for aerial tracking that eliminates ESDF updates and incorporates a recovery-capable path generation method for target reacquisition. First, we design a target trajectory prediction method and a visibility-aware initial path generation algorithm that maintain an appropriate observation distance, avoid occlusions, and enable rapid replanning to reacquire the target when it is lost. Then, we propose the Field of View ESDF (FoV-ESDF), a precomputed ESDF tailored to the tracker's field of view, enabling rapid visibility evaluation without requiring updates. Finally, we optimize the trajectory using differentiable FoV-ESDF-based objectives to ensure continuous visibility throughout the tracking process. Extensive simulations and real-world experiments demonstrate that our approach delivers more robust tracking results with lower computational effort than existing state-of-the-art methods. The source code is available at https://github.com/Yue-0/Eva-Tracker.

Figures (9)

• Figure 1: Comparison between the traditional Euclidean Signed Distance Field (ESDF) and the proposed Field of View ESDF (FoV-ESDF). (a) The traditional ESDF encodes the distance from each voxel to its nearest obstacle across a wide spatial range. It must be updated in real time to reflect environmental changes, resulting in high computational overhead and substantial memory usage. (b) A slice of the FoV-ESDF, with white lines indicating the boundaries of the tracker's FoV. The FoV-ESDF only encodes distances from voxels within the FoV to their nearest FoV boundary, while ignoring voxels outside the FoV. It is constructed once, requires significantly less memory, and eliminates the need for subsequent updates.
• Figure 2: Overview of the proposed system and the real-world drone platform. Left: The perception module processes sensor data for target detection, localization, and local mapping, all of which are executed entirely on the drone's onboard computing device. Right: The trajectory planning module comprises three core components: target trajectory prediction, visibility-aware initial path generation, and FoV-ESDF-guided trajectory optimization. These components collaborate to produce occlusion-free tracking trajectories, ensuring continuous target visibility and enabling recovery when the target is lost.
• Figure 3: Illustration of visibility-aware initial path generation.
• Figure 4: Illustration of FoV-ESDF-guided visibility optimization. Green circles represent obstacles within the FoV, while green arrows indicate the direction of the gradient descent generated by the FoV-ESDF for these obstacles. A red arrow indicates the direction of the gradient descent generated by the FoV-ESDF for the target. Orange circles represent obstacles that do not enter the FoV, which are excluded from the calculation.
• Figure 5: The random simulation environment and the target's trajectory.