Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Target Radar Search and Track Using Sequence-Capable Deep Reinforcement Learning

Jan-Hendrik Ewers, David Cormack, Joe Gibbs, David Anderson

TL;DR

The paper addresses radar sensor management for simultaneous search and multi-target tracking (MTT) in a full 3D environment by developing a 3D simulation with an AESA radar and comparing three sequence-capable neural architectures, augmented with Behaviour Cloning and Auto-Encoder pretraining. It demonstrates that MHSA-enabled sequence models (PPO-BiGRU+MHSA) best balance search and tracking performance, while static policies lag due to lack of tracking. Key contributions include showing the effectiveness of sequence-capable architectures in dynamic MTT tasks and providing a framework that balances search quality and tracking fidelity through a principled reward structure. The findings have practical implications for optimizing radar sensor management and suggest avenues for future enhancements, such as incorporating platform motion and richer trajectory histories.

Abstract

The research addresses sensor task management for radar systems, focusing on efficiently searching and tracking multiple targets using reinforcement learning. The approach develops a 3D simulation environment with an active electronically scanned array radar, using a multi-target tracking algorithm to improve observation data quality. Three neural network architectures were compared including an approach using fated recurrent units with multi-headed self-attention. Two pre-training techniques were applied: behavior cloning to approximate a random search strategy and an auto-encoder to pre-train the feature extractor. Experimental results revealed that search performance was relatively consistent across most methods. The real challenge emerged in simultaneously searching and tracking targets. The multi-headed self-attention architecture demonstrated the most promising results, highlighting the potential of sequence-capable architectures in handling dynamic tracking scenarios. The key contribution lies in demonstrating how reinforcement learning can optimize sensor management, potentially improving radar systems' ability to identify and track multiple targets in complex environments.

Multi-Target Radar Search and Track Using Sequence-Capable Deep Reinforcement Learning

TL;DR

The paper addresses radar sensor management for simultaneous search and multi-target tracking (MTT) in a full 3D environment by developing a 3D simulation with an AESA radar and comparing three sequence-capable neural architectures, augmented with Behaviour Cloning and Auto-Encoder pretraining. It demonstrates that MHSA-enabled sequence models (PPO-BiGRU+MHSA) best balance search and tracking performance, while static policies lag due to lack of tracking. Key contributions include showing the effectiveness of sequence-capable architectures in dynamic MTT tasks and providing a framework that balances search quality and tracking fidelity through a principled reward structure. The findings have practical implications for optimizing radar sensor management and suggest avenues for future enhancements, such as incorporating platform motion and richer trajectory histories.

Abstract

The research addresses sensor task management for radar systems, focusing on efficiently searching and tracking multiple targets using reinforcement learning. The approach develops a 3D simulation environment with an active electronically scanned array radar, using a multi-target tracking algorithm to improve observation data quality. Three neural network architectures were compared including an approach using fated recurrent units with multi-headed self-attention. Two pre-training techniques were applied: behavior cloning to approximate a random search strategy and an auto-encoder to pre-train the feature extractor. Experimental results revealed that search performance was relatively consistent across most methods. The real challenge emerged in simultaneously searching and tracking targets. The multi-headed self-attention architecture demonstrated the most promising results, highlighting the potential of sequence-capable architectures in handling dynamic tracking scenarios. The key contribution lies in demonstrating how reinforcement learning can optimize sensor management, potentially improving radar systems' ability to identify and track multiple targets in complex environments.

Paper Structure

This paper contains 12 sections, 1 equation, 1 figure.

Table of Contents

  1. Introduction
  2. Methodology
  3. Simulation and MTT Feature Extraction
  4. Scan History
  5. Action Space
  6. Observation Space
  7. Rewards
  8. Network Design
  9. Behaviour Cloning
  10. Auto-encoder
  11. Results
  12. Conclusions and Future Work

Figures (1)

  • Figure 1: GOSPA distance (\ref{['fig:results_gospa_distance']}) and three of its components Rahmathullah2017. GOSPA switching has been excluded as algorithms had a value of $0$ for this metric showing that any detected track was never falsely associated to a different ground truth throughout its existence.