Physics-aware Learnable State Space Model for UAV Trajectory Prediction

Read full text Summarise with VIRAT
Authors: Xiaofeng Zhang
Cited by: 0
References: 11

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Unmanned aerial vehicle (UAV) trajectory prediction plays an important role in autonomous flight control, path planning, and obstacle avoidance. However, long-horizon autoregressive prediction often suffers from error accumulation, limited physical consistency, and reduced robustness under highly dynamic maneuvers. To address these challenges, this paper proposes a Physics-aware Learnable State Space Predictor (PLSSP). Instead of directly regressing future positions, the proposed framework represents UAV motion using a joint physical state composed of position, velocity, and acceleration. A residual dynamics learning strategy is introduced to model only the dynamic perturbation component, while an input-driven selective state update mechanism enables adaptive state evolution across heterogeneous flight regimes. Furthermore, a stability-oriented error propagation analysis is developed to characterize recursive prediction behavior. Experiments on four physics-inspired synthetic UAV trajectory benchmarks with progressively increasing maneuver complexity demonstrate that PLSSP achieves competitive prediction accuracy, improved physical consistency under aggressive maneuvers, and lower absolute error accumulation during long-horizon prediction.

Fields of Study & Subjects

Aerospace and Aviation Technology Aerospace Engineering and Control Systems Reinforcement Learning in Robotics

Publication details

Journal / source
Journal of Intelligent & Robotic Systems
Publication date
12 November 2026
Language
EN