RaaS oriented semantic task management framework for heterogeneous multi-robot systems

다국어 초록 (Multilingual Abstract)

To achieve autonomous collaboration among Heterogeneous Multi-Robot Systems in complex indoor environments, a unified control architecture capable of overcoming physical heterogeneity and managing the semantic context of spaces and missions is essential. This paper proposes a RaaS (Robot as a Service)-oriented Semantic Task Management Framework that normalizes robot, space, and task information into a three-layer structure (Symbolic, Explicit, Implicit) utilizing the Triplet Ontological Semantic Model (TOSM).
The proposed framework comprises four core modules. A Semantic Map Generator that enhances semantic completeness by integrating geometric exploration with semantic information gain. An LLM-based Semantic Task Planner that converts and grounds unstructured natural language commands into executable tasks based on TOSM. A Cost-aware Task Allocation algorithm that comprehensively evaluates robot capabilities, operational costs, and priorities. and a Failure-Aware Adaptive Reallocation mechanism that responds in real-time to runtime execution failures. Simulation experiments conducted in hospital and office environments demonstrated that the proposed method achieves high accuracy in interpreting complex natural language commands and efficiency in task allocation. Furthermore, the system verified its robustness through rapid reallocation in dynamic failure situations. This study contributes to substantially enhancing the interoperability and service continuity of heterogeneous robot fleets by integrating semantic technologies with the RaaS architecture.

국문 초록 (Abstract)

본 논문은 Triplet Ontological Semantic Model을 기반으로 로봇, 공간, 작업정보를 Symbolic, Explicit, Implicit의 3개의 계층으로 정규화하고, 이를 서버 환경에서 통합 운용하는 RaaS(Robot as a Service) 지향형 의미론적 작업 관리 프레임워크를 제안한다.
제안하는 프레임워크는 기하학적 탐색과 의미론적 정보 이득을 결합하여 환경의 의미적 완성도를 높이는 Semantic Map Generator, 대형 언어 모델(LLM)을 활용해 비정형 자연어 명령을 TOSM 기반의 실행 가능한 태스크로 변환 및 접지하는 의미론적 작업 계획기, 로봇의 기능과 운용 비용 및 우선순위를 종합적으로 평가하는 비용 기반 작업할당, 그리고 실패 상황에 실시간으로 대응하는 메커니즘으로 구성된다. 시뮬레이션 실험을 통해, 제안 기법은 복합 자연어 명령의 높은 해석 정확도와 효율적인 작업 할당 성능을 보였으며, 동적 실패 상황에서의 신속한 재할당을 통해 시스템의 강건성을 입증하였다. 본 연구는 Semantic Task Management, RaaS 아키텍처의 결합을 통해 이기종 로봇 플릿의 상호운용성과 서비스 연속성을 실질적으로 향상시키는 데 기여한다.

목차 (Table of Contents)

• 1. Introduction 1
• 2. Related Works 7
• 2.1. SLAM and Autonomous Exploration 7
• 2.2. Semantic SLAM and Map Expression 10
• 2.3. LLM-based Multi Robot Task Management 13
• 2.4. RaaS-oriented Heterogeneous Multi-Robot Management 15
• 3. Methodology 18
• 3.1. Semantic Map Generator Method 18
• 3.1.1. Semantic Information–Driven Autonomous Exploration 19
• 3.1.2. Room Segmentation based on Voronoi Graph 26
• 3.2. Semantic Task Manager 28
• 3.2.1. Robot Triplet Ontological Semantic Modeling 28
• 3.2.2. Task Triplet Ontological Semantic Modeling 31
• 3.2.3. LLM-based Semantic Task Planner 34
• 3.2.4. Task–Robot Matching based on Semantic DB 38
• 3.2.5. Robot State Validation 41
• 3.2.6. Cost-aware Heterogeneous Multi-Robot Task Allocation 44
• 3.2.7. Failure-Aware Adaptive Task Reallocation 47
• 3.3. RaaS-oriented Robot Control Server System 50
• 3.3.1. System Architecture of the RaaS Control Server 51
• 3.3.2. Communication Architecture of the RaaS Control Server 53
• 3.3.3. Operational Architecture of the RaaS Control Server 54
• 4. Experiments and Results 58
• 4.1. Experiment Setup 58
• 4.1.1 Robot Semantic Database Modeling 59
• 4.1.2 Environment Semantic Database Modeling 61
• 4.1.2.1 Hospital Environment 63
• 4.1.2.2 Office Environment 66
• 4.1.3. Evaluation Metrics 68
• 4.2.1. Scenario 1: Complex Sequential Command Execution 72
• 4.2.2. Scenario 2: Sensitivity Analysis of Cost-aware Task Allocation 75
• 4.2.2.1. Scenario 2-1: Parallel Batch Allocation 76
• 4.2.2.2. Scenario 2-2: Priority-aware Sequential Allocation 78
• 4.2.3. Scenario 3: Robot Failure Handling and Adaptive Reallocation 80
• 5. Conclusion and Future Works 85
• 5.1. Conclusion 85
• 5.2. Future Works 87
• References 91
• Korean Abstract 99