Distributed Online Task Assignment via Inexact ADMM for unplanned online tasks and its Applications to Security

2025-02-26

AI Summary

To address the challenge of simultaneously ensuring safety and task coordination in multi-robot systems under concurrent threats of emergent online tasks and adversarial plan-deviation attacks, this paper proposes a distributed task allocation framework. The method innovatively integrates inexact Alternating Direction Method of Multipliers (ADMM) with a safety-aware online decision-making mechanism, and— for the first time— embeds the Control Lyapunov Function–Control Barrier Function (CLF-CBF) joint control paradigm into the distributed task execution loop, enabling real-time co-enforcement of task fulfillment and safety constraints. By synergistically optimizing projection-based gradient descent and a formal safety analysis model, the framework is validated under communication-constrained conditions: it achieves 100% safety constraint satisfaction, improves task allocation convergence speed by 40%, and demonstrates significantly enhanced robustness compared to centralized baseline approaches.

Technology Category

• Multiagent Systems: Adversarial Agents
• Intelligent Robots: Multi-Robot Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

In multi-robot system (MRS) applications, efficient task assignment is essential not only for coordinating agents and ensuring mission success but also for maintaining overall system security. In this work, we first propose an optimization-based distributed task assignment algorithm that dynamically assigns mandatory security-critical tasks and optional tasks among teams. Leveraging an inexact Alternating Direction Method of Multipliers (ADMM)-based approach, we decompose the task assignment problem into separable and non-separable subproblems. The non-separable subproblems are transformed into an inexact ADMM update by projected gradient descent, which can be performed through several communication steps within the team. In the second part of this paper, we formulate a comprehensive framework that enables MRS under plan-deviation attacks to handle online tasks without compromising security. The process begins with a security analysis that determines whether an online task can be executed securely by a robot and, if so, the required time and location for the robot to rejoin the team. Next, the proposed task assignment algorithm is used to allocate security-related tasks and verified online tasks. Finally, task fulfillment is managed using a Control Lyapunov Function (CLF)-based controller, while security enforcement is ensured through a Control Barrier Function (CBF)-based security filter. Through simulations, we demonstrate that the proposed framework allows MRS to effectively respond to unplanned online tasks while maintaining security guarantees.

Problem

• Dynamic assignment of security-critical tasks in MRS.
• Handling online tasks during plan-deviation attacks securely.
• Decomposing task assignment using inexact ADMM approach.

Innovation

• Inexact ADMM for distributed task assignment
• Security analysis for online task execution
• CLF and CBF controllers for task and security