Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Control-Aware Trajectory Predictions for Communication-Efficient Drone Swarm Coordination in Cluttered Environments

Longhao Yan, Jingyuan Zhou, Kaidi Yang

TL;DR

This work tackles safe, communication-efficient UAV swarm coordination in cluttered environments by predicting neighbors' planned trajectories under limited bandwidth. It introduces a control-aware predictor that fuses an EvolveGCN-based dynamic-graph model with a VAE-based compressor, trained with a KKT-informed objective to align predictions with the embedded DMPC optimization. A Bayesian fusion integrates EG and VAE outputs to generate robust neighbor trajectories, enabling near-optimal DMPC performance despite communication constraints. Empirical results in a funnel-like setting show the approach outperforms baselines and maintains safety under reduced communication frequency and measurement noise. Overall, the method advances scalable, decentralized swarm control with practical communication limitations.

Abstract

Swarms of Unmanned Aerial Vehicles (UAV) have demonstrated enormous potential in many industrial and commercial applications. However, before deploying UAVs in the real world, it is essential to ensure they can operate safely in complex environments, especially with limited communication capabilities. To address this challenge, we propose a control-aware learning-based trajectory prediction algorithm that can enable communication-efficient UAV swarm control in a cluttered environment. Specifically, our proposed algorithm can enable each UAV to predict the planned trajectories of its neighbors in scenarios with various levels of communication capabilities. The predicted planned trajectories will serve as input to a distributed model predictive control (DMPC) approach. The proposed algorithm combines (1) a trajectory prediction model based on EvolveGCN, a Graph Convolutional Network (GCN) that can handle dynamic graphs, which is further enhanced by compressed messages from adjacent UAVs, and (2) a KKT-informed training approach that applies the Karush-Kuhn-Tucker (KKT) conditions in the training process to encode DMPC information into the trained neural network. We evaluate our proposed algorithm in a funnel-like environment. Results show that the proposed algorithm outperforms state-of-the-art benchmarks, providing close-to-optimal control performance and robustness to limited communication capabilities and measurement noises.

Control-Aware Trajectory Predictions for Communication-Efficient Drone Swarm Coordination in Cluttered Environments

TL;DR

This work tackles safe, communication-efficient UAV swarm coordination in cluttered environments by predicting neighbors' planned trajectories under limited bandwidth. It introduces a control-aware predictor that fuses an EvolveGCN-based dynamic-graph model with a VAE-based compressor, trained with a KKT-informed objective to align predictions with the embedded DMPC optimization. A Bayesian fusion integrates EG and VAE outputs to generate robust neighbor trajectories, enabling near-optimal DMPC performance despite communication constraints. Empirical results in a funnel-like setting show the approach outperforms baselines and maintains safety under reduced communication frequency and measurement noise. Overall, the method advances scalable, decentralized swarm control with practical communication limitations.

Abstract

Swarms of Unmanned Aerial Vehicles (UAV) have demonstrated enormous potential in many industrial and commercial applications. However, before deploying UAVs in the real world, it is essential to ensure they can operate safely in complex environments, especially with limited communication capabilities. To address this challenge, we propose a control-aware learning-based trajectory prediction algorithm that can enable communication-efficient UAV swarm control in a cluttered environment. Specifically, our proposed algorithm can enable each UAV to predict the planned trajectories of its neighbors in scenarios with various levels of communication capabilities. The predicted planned trajectories will serve as input to a distributed model predictive control (DMPC) approach. The proposed algorithm combines (1) a trajectory prediction model based on EvolveGCN, a Graph Convolutional Network (GCN) that can handle dynamic graphs, which is further enhanced by compressed messages from adjacent UAVs, and (2) a KKT-informed training approach that applies the Karush-Kuhn-Tucker (KKT) conditions in the training process to encode DMPC information into the trained neural network. We evaluate our proposed algorithm in a funnel-like environment. Results show that the proposed algorithm outperforms state-of-the-art benchmarks, providing close-to-optimal control performance and robustness to limited communication capabilities and measurement noises.
Paper Structure (12 sections, 1 theorem, 13 equations, 5 figures, 10 tables)

This paper contains 12 sections, 1 theorem, 13 equations, 5 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Problem Statement
  3. Control-aware learning-based trajectory prediction
  4. Trajectory predictor
  5. KKT-informed training of the trajectory prediction neural network
  6. Simulation and Analysis
  7. Benchmarks
  8. Implementation details
  9. Performance in testing scenarios
  10. Robustness in scenarios with limited communication capabilities
  11. Sensitivity to measurement noises
  12. Conclusion

Key Result

Proposition 1

Given primal solution $\bm{w}^*$ and dual solution $\bm{\lambda}^*$ to the following QP where the parameters are functions with respect to parameters $\bm{u}$. Then the derivatives of the primal and dual solutions $(\frac{\partial \bm{w}^*}{\partial \bm{u}^*}, \frac{\partial \bm{\lambda}^*}{\partial \bm{u}^*})$ satisfy

Figures (5)

  • Figure 1: Illustration of the considered scenario, where the red point represents the goal of the swarm, and the green ellipsoids represent the obstacles.
  • Figure 2: Framework of the trajectory prediction algorithm
  • Figure 3: Example of KKT-informed training process, whereby UAV 1 wants to estimate the planned trajectories of UAV 2. UAV 1, 3, 4, and 5 are the closest neighbors of UAV 2.
  • Figure 4: Safe distance comparison of different benchmarks.
  • Figure 5: Evolution of distances with different benchmarks applied.

Theorems & Definitions (1)

  • Proposition 1: Differentiable KKT Layer amos2017optnet