Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Grasper: A Generalist Pursuer for Pursuit-Evasion Problems

Pengdeng Li, Shuxin Li, Xinrun Wang, Jakub Cerny, Youzhi Zhang, Stephen McAleer, Hau Chan, Bo An

TL;DR

Grasper tackles the challenge of solving pursuit-evasion problems with varying initial conditions by conditioning pursuer policies on PEG-specific graphs. It combines a graph neural encoder with a hypernetwork to generate a PEG-conditioned base policy, and employs a three-stage training pipeline—GraphMAE-based pre-pretraining, heuristic-guided multi-task pre-training, and PSRO-based fine-tuning—to achieve fast, generalizable BR policy learning. Across synthetic and real-world maps, Grasper outperforms strong baselines in both solution quality and generalizability, while also stabilizing training through effective pre-training components. This approach enables practical deployment of pursuer policies that adapt to diverse and dynamic urban PEG scenarios. It represents a substantive advance in generalizing PEG solutions beyond fixed initial conditions, with potential impact on real-time security resource allocation and automated pursuit planning.

Abstract

Pursuit-evasion games (PEGs) model interactions between a team of pursuers and an evader in graph-based environments such as urban street networks. Recent advancements have demonstrated the effectiveness of the pre-training and fine-tuning paradigm in PSRO to improve scalability in solving large-scale PEGs. However, these methods primarily focus on specific PEGs with fixed initial conditions that may vary substantially in real-world scenarios, which significantly hinders the applicability of the traditional methods. To address this issue, we introduce Grasper, a GeneRAlist purSuer for Pursuit-Evasion pRoblems, capable of efficiently generating pursuer policies tailored to specific PEGs. Our contributions are threefold: First, we present a novel architecture that offers high-quality solutions for diverse PEGs, comprising critical components such as (i) a graph neural network (GNN) to encode PEGs into hidden vectors, and (ii) a hypernetwork to generate pursuer policies based on these hidden vectors. As a second contribution, we develop an efficient three-stage training method involving (i) a pre-pretraining stage for learning robust PEG representations through self-supervised graph learning techniques like GraphMAE, (ii) a pre-training stage utilizing heuristic-guided multi-task pre-training (HMP) where heuristic-derived reference policies (e.g., through Dijkstra's algorithm) regularize pursuer policies, and (iii) a fine-tuning stage that employs PSRO to generate pursuer policies on designated PEGs. Finally, we perform extensive experiments on synthetic and real-world maps, showcasing Grasper's significant superiority over baselines in terms of solution quality and generalizability. We demonstrate that Grasper provides a versatile approach for solving pursuit-evasion problems across a broad range of scenarios, enabling practical deployment in real-world situations.

Grasper: A Generalist Pursuer for Pursuit-Evasion Problems

TL;DR

Grasper tackles the challenge of solving pursuit-evasion problems with varying initial conditions by conditioning pursuer policies on PEG-specific graphs. It combines a graph neural encoder with a hypernetwork to generate a PEG-conditioned base policy, and employs a three-stage training pipeline—GraphMAE-based pre-pretraining, heuristic-guided multi-task pre-training, and PSRO-based fine-tuning—to achieve fast, generalizable BR policy learning. Across synthetic and real-world maps, Grasper outperforms strong baselines in both solution quality and generalizability, while also stabilizing training through effective pre-training components. This approach enables practical deployment of pursuer policies that adapt to diverse and dynamic urban PEG scenarios. It represents a substantive advance in generalizing PEG solutions beyond fixed initial conditions, with potential impact on real-time security resource allocation and automated pursuit planning.

Abstract

Pursuit-evasion games (PEGs) model interactions between a team of pursuers and an evader in graph-based environments such as urban street networks. Recent advancements have demonstrated the effectiveness of the pre-training and fine-tuning paradigm in PSRO to improve scalability in solving large-scale PEGs. However, these methods primarily focus on specific PEGs with fixed initial conditions that may vary substantially in real-world scenarios, which significantly hinders the applicability of the traditional methods. To address this issue, we introduce Grasper, a GeneRAlist purSuer for Pursuit-Evasion pRoblems, capable of efficiently generating pursuer policies tailored to specific PEGs. Our contributions are threefold: First, we present a novel architecture that offers high-quality solutions for diverse PEGs, comprising critical components such as (i) a graph neural network (GNN) to encode PEGs into hidden vectors, and (ii) a hypernetwork to generate pursuer policies based on these hidden vectors. As a second contribution, we develop an efficient three-stage training method involving (i) a pre-pretraining stage for learning robust PEG representations through self-supervised graph learning techniques like GraphMAE, (ii) a pre-training stage utilizing heuristic-guided multi-task pre-training (HMP) where heuristic-derived reference policies (e.g., through Dijkstra's algorithm) regularize pursuer policies, and (iii) a fine-tuning stage that employs PSRO to generate pursuer policies on designated PEGs. Finally, we perform extensive experiments on synthetic and real-world maps, showcasing Grasper's significant superiority over baselines in terms of solution quality and generalizability. We demonstrate that Grasper provides a versatile approach for solving pursuit-evasion problems across a broad range of scenarios, enabling practical deployment in real-world situations.
Paper Structure (20 sections, 5 figures, 2 tables, 3 algorithms)

This paper contains 20 sections, 5 figures, 2 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Preliminaries
  5. Policy-Space Response Oracles
  6. Problem Statement
  7. Grasper
  8. Architecture
  9. Graphical Representations of PEGs.
  10. Game-conditional Base Policies Generation.
  11. Observation Representation Layer.
  12. Training Pipeline
  13. Stage I: Pre-pretraining.
  14. Stage II: Pre-training.
  15. Stage III: Fine-tuning.
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Architecture and training pipeline of Grasper.
  • Figure 2: Illustration of HMP.
  • Figure 3: Evaluation performance. The shaded area represents the standard error.
  • Figure 4: Pre-training curves.
  • Figure 5: Zero-shot worst-case utility of the pursuer for each possible evader's initial location. Red dots are exits and blue dots are pursuers' initial locations.