ROAR a Robust Autonomous Aerial Tracking System for Challenging Scenarios
Tong Zhang, Chenghao Li, Kezhen Zhao, Hao Shen, Tao Pang
AI summary
Problem
Autonomous UAV tracking frequently fails in complex environments due to target loss from occlusion or FOV exit, while existing trajectory optimization methods often suffer from high computational burden and poor yaw control.
Approach
The ROAR system predicts target positions using a Markov chain to generate potential viewpoints and triggers a re-capture strategy when tracking is lost, while a B-spline trajectory optimizer jointly minimizes tracking distance and optimizes yaw angle for safe, smooth flight.
Key results
- Markov chain motion prediction framework with heuristic viewpoint generation
- Re-capture strategy for efficient target recovery after FOV loss
- Joint B-spline trajectory optimizer with independent yaw and time-dependent distance costs
- Validated superior tracking success rates and FOV retention in simulations and real-world tests
Why it matters
Provides a robust, computationally efficient tracking solution for UAVs operating in cluttered or dynamic environments where traditional methods struggle.
Abstract
Autonomous tracking represents a significant ad- vancement in the evolution of unmanned aerial vehicles (UAVs), offering applications in areas such as aerial photography and infrastructure inspection. Despite its potential, many autonomous tracking systems encounter challenges in maintaining consistent and reliable target tracking, particularly in complex and dynamic environments. This study presents a robust approach to au- tonomous tracking aimed at overcoming two primary challenges that often lead to tracking failures: target loss and poor tracking trajectory quality. To tackle these issues, a Markov chain- based motion prediction method is introduced to estimate the target’s future position probabilities over time. Based on these predictions, a re-capture strategy is designed to enhance decision- making, ensuring effective recovery when the target exits the field of view (FOV). Additionally, the tracking process is modified by formulating cost functions that incorporate tracking distance and optimal yaw angle, while ensuring safety, dynamic feasibility, and operational constraints. Simulations and real-world experiments validate the proposed method, demonstrating stable and efficient tracking performance in challenging environments.