Drones Help Drones: A Collaborative Framework for Multi-Drone Object Trajectory Prediction and Beyond

TL;DR

This work tackles multi-drone object trajectory prediction under communication constraints by introducing Drones Help Drones (DHD), a framework that combines a Ground-prior-based BEV Generation (GBG) module with a Sparse Interaction via Sliding Windows (SISW) module to create accurate BEV representations while curbing data transmission. The approach estimates depth more reliably in aerial views by leveraging ground intersections and object height, then selectively exchanges information from high-discrepancy regions to maintain real-time performance. A temporal U-Net decoder jointly predicts future segmentation and motion, trained with a cross-entropy and an $L_1$ flow loss under a temporal discount, and evaluated on the Air-Co-Pred dataset, a first of its kind for multi-drone collaborative trajectory prediction. Results show over 20% reduction in BEV position deviation and a 4x reduction in inter-drone transmission with competitive or superior prediction accuracy, alongside demonstrated generalization to collaborative 3D object detection in CoPerception-UAVs. The work advances bandwidth-aware collaboration for aerial perception with practical implications for safe, efficient multi-drone operations in complex environments.

Abstract

Collaborative trajectory prediction can comprehensively forecast the future motion of objects through multi-view complementary information. However, it encounters two main challenges in multi-drone collaboration settings. The expansive aerial observations make it difficult to generate precise Bird's Eye View (BEV) representations. Besides, excessive interactions can not meet real-time prediction requirements within the constrained drone-based communication bandwidth. To address these problems, we propose a novel framework named "Drones Help Drones" (DHD). Firstly, we incorporate the ground priors provided by the drone's inclined observation to estimate the distance between objects and drones, leading to more precise BEV generation. Secondly, we design a selective mechanism based on the local feature discrepancy to prioritize the critical information contributing to prediction tasks during inter-drone interactions. Additionally, we create the first dataset for multi-drone collaborative prediction, named "Air-Co-Pred", and conduct quantitative and qualitative experiments to validate the effectiveness of our DHD framework.The results demonstrate that compared to state-of-the-art approaches, DHD reduces position deviation in BEV representations by over 20% and requires only a quarter of the transmission ratio for interactions while achieving comparable prediction performance. Moreover, DHD also shows promising generalization to the collaborative 3D object detection in CoPerception-UAVs.

Figures (11)

• Figure 1: Comparative visualization of observations in autonomous driving versus aerial surveillance.
• Figure 2: The overall architecture of our proposed DHD framework. For clarity, we just present the collaboration between two drones.
• Figure 3: Illustration of the theoretical depth upper-bound and the impact of various viewing angles on depth estimation.The depth of objects, $D_1$, can not exceed $D_\text{upperbound}$.
• Figure 4: Operation flow of GBG module.
• Figure 5: Operation flow of SISW Module.