Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Perception-to-Pursuit: Track-Centric Temporal Reasoning for Open-World Drone Detection and Autonomous Chasing

Venkatakrishna Reddy Oruganti

TL;DR

This work targets the perception-to-pursuit gap in autonomous counter-drone systems by introducing Perception-to-Pursuit (P2P), a track-centric temporal reasoning framework that predicts motion-aware trajectories feasible for interception. It achieves this with an 8-D motion token representation and a 12-frame causal transformer that jointly predicts drone class, behavior, intent, and a future trajectory, constrained by kinematic reachability. A new Intercept Success Rate (ISR) metric directly quantifies pursuit feasibility under realistic interceptor limits, and experiments on Anti-UAV-RGBT show a 77% ADE improvement and 597x improvement in ISR over tracking-only baselines, while maintaining 100% classification accuracy. The results demonstrate that temporal motion reasoning yields both accurate trajectory forecasting and practically actionable pursuit plans, with real-time performance suitable for deployment.

Abstract

Autonomous drone pursuit requires not only detecting drones but also predicting their trajectories in a manner that enables kinematically feasible interception. Existing tracking methods optimize for prediction accuracy but ignore pursuit feasibility, resulting in trajectories that are physically impossible to intercept 99.9% of the time. We propose Perception-to-Pursuit (P2P), a track-centric temporal reasoning framework that bridges detection and actionable pursuit planning. Our method represents drone motion as compact 8-dimensional tokens capturing velocity, acceleration, scale, and smoothness, enabling a 12-frame causal transformer to reason about future behavior. We introduce the Intercept Success Rate (ISR) metric to measure pursuit feasibility under realistic interceptor constraints. Evaluated on the Anti-UAV-RGBT dataset with 226 real drone sequences, P2P achieves 28.12 pixel average displacement error and 0.597 ISR, representing a 77% improvement in trajectory prediction and 597x improvement in pursuit feasibility over tracking-only baselines, while maintaining perfect drone classification accuracy (100%). Our work demonstrates that temporal reasoning over motion patterns enables both accurate prediction and actionable pursuit planning.

Perception-to-Pursuit: Track-Centric Temporal Reasoning for Open-World Drone Detection and Autonomous Chasing

TL;DR

This work targets the perception-to-pursuit gap in autonomous counter-drone systems by introducing Perception-to-Pursuit (P2P), a track-centric temporal reasoning framework that predicts motion-aware trajectories feasible for interception. It achieves this with an 8-D motion token representation and a 12-frame causal transformer that jointly predicts drone class, behavior, intent, and a future trajectory, constrained by kinematic reachability. A new Intercept Success Rate (ISR) metric directly quantifies pursuit feasibility under realistic interceptor limits, and experiments on Anti-UAV-RGBT show a 77% ADE improvement and 597x improvement in ISR over tracking-only baselines, while maintaining 100% classification accuracy. The results demonstrate that temporal motion reasoning yields both accurate trajectory forecasting and practically actionable pursuit plans, with real-time performance suitable for deployment.

Abstract

Autonomous drone pursuit requires not only detecting drones but also predicting their trajectories in a manner that enables kinematically feasible interception. Existing tracking methods optimize for prediction accuracy but ignore pursuit feasibility, resulting in trajectories that are physically impossible to intercept 99.9% of the time. We propose Perception-to-Pursuit (P2P), a track-centric temporal reasoning framework that bridges detection and actionable pursuit planning. Our method represents drone motion as compact 8-dimensional tokens capturing velocity, acceleration, scale, and smoothness, enabling a 12-frame causal transformer to reason about future behavior. We introduce the Intercept Success Rate (ISR) metric to measure pursuit feasibility under realistic interceptor constraints. Evaluated on the Anti-UAV-RGBT dataset with 226 real drone sequences, P2P achieves 28.12 pixel average displacement error and 0.597 ISR, representing a 77% improvement in trajectory prediction and 597x improvement in pursuit feasibility over tracking-only baselines, while maintaining perfect drone classification accuracy (100%). Our work demonstrates that temporal reasoning over motion patterns enables both accurate prediction and actionable pursuit planning.
Paper Structure (22 sections, 16 equations, 2 figures, 3 tables)

This paper contains 22 sections, 16 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Drone Detection and Tracking
  4. Trajectory Prediction
  5. Pursuit and Interception
  6. Temporal Reasoning
  7. Method
  8. Problem Formulation
  9. Motion Token Representation
  10. Temporal Transformer Architecture
  11. Training Objective
  12. Intercept Success Rate Metric
  13. Experiments
  14. Implementation Details
  15. Quantitative Results
  16. ...and 7 more sections

Figures (2)

  • Figure 1: Perception-to-Pursuit Framework. Our pipeline processes tracked bounding boxes into 8-dimensional motion tokens capturing velocity, acceleration, scale, and smoothness. A 12-frame causal transformer reasons over these tokens to jointly predict: (1) drone classification, (2) behavior category, (3) maneuver intent, and (4) future trajectory. Multi-task learning ensures predictions are both accurate and pursuit-feasible.
  • Figure 2: Qualitative examples. Top: Success case with complex maneuver. Our method (green) accurately predicts evasive trajectory while tracking-only (red) produces infeasible linear extrapolation. Bottom: Challenging case with rapid direction change. Our prediction enables pursuit (ISR=0.78) while baseline fails (ISR=0.02).