Multi-Agent Simultaneous Segmentation and Scheduling

Multi-Agent Simultaneous Segmentation and Scheduling (MASSS), a transformative approach aimed at enhancing the capabilities and flexibility of Autonomous Mobile Manipulators (AMMs) within depot and defense manufacturing processes. By improving AMMs' ability to initially perform discrete and eventually continuous tasks across heterogeneous robotic systems, MASSS will optimize efficiency, adaptability, and resource utilization. However, solving the complex challenges associated with MASSS requires a focused effort, and we aim to address the most critical technical needs while ensuring end-to-end system functionality. Current state-of-the-art systems typically rely on a manual, iterative process to generate open-loop strategies for task allocation, often requiring a robotic expert and a week or more to complete. MASSS addresses this inefficiency by enabling continuous replanning of tasks as conditions evolve, allowing robots to adapt to variability in process execution. This adaptive capability contrasts with open-loop solutions, which lack the flexibility to handle dynamic changes. To meet the core needs of MASSS, this project will focus on three key areas: task modeling and process estimation, task assignment and optimization, and simulation and preemptive replanning. We will build a stand-alone application capable of controlling heterogeneous robotic systems, providing a proof of concept and a tool for real-time exploration of the impact of adding or removing robots from the worksite. Expanding this tool to include task time estimators and feedback-learning modules will be a secondary goal if time permits. The task modeling and process estimation effort will be tackled in three phases: specifying the outputs of modeling components, automating the generation of this data, and incorporating machine learning to adapt to real-world data. The task scheduling and optimization will focus on developing efficient algorithms to evaluate and explore various task assignments and schedules. Finally, a low-fidelity simulation of the MASSS system will validate algorithm performance and preemptive replanning capabilities, serving as a versatile tool for both testing and decision-making.