文献类型：期刊文献

英文题名：Constraint-Aware Multi-Agent Proximal Policy Optimization for Online Multi-Droplet Scheduling in Digital Microfluidics

作者：Guo, Kunlun[1]; Song, Zerui[1]; Wang, Huifeng[1]

机构：[1] East China University of Science and Technology, School of Information Science and Engineering, Shanghai, China

年份：2026

外文期刊名：Proceeding: ECIS 2026 - 2026 IEEE 3rd International Conference on Electronics, Communications and Intelligent Science

收录：EI(收录号：20263221245827)

语种：英文

外文关键词：Constrained optimization - Decision making - Drop breakup - Drop formation - Electrodes - Intelligent agents - Safety engineering

摘要：Online scheduling of multiple droplets in digital microfluidics (DMF) is challenged by frequent path conflicts, strict inter-droplet safe-spacing constraints, and local coordination failures on limited electrode arrays. To address these issues, this paper proposes a constraint-aware multi-agent proximal policy optimization method, termed CA-MAPPO. First, the DMF electrode array is modeled as a grid graph, and the inter-droplet physical constraints, including unique electrode occupancy and Chebyshev-distance-based safe spacing, are explicitly associated with a multi-agent sequential decision-making formulation. In this formulation, each droplet is represented as an agent, while local observations and global states encode droplet positions, task states, and safety-related spatial information under the centralized training with decentralized execution framework. Furthermore, safety-constrained policy optimization is introduced to mask infeasible actions before sampling, and priorityguided conflict resolution is designed to coordinate competing droplets in local bottleneck regions. Simulation experiments on a 5 × 18 electrode array show that CA-MAPPO achieves a success rate of 90.3%, an average of 22.7 scheduling steps, and 2.42 unsafe candidate actions across scenarios with 3 to 6 droplets. Compared with online greedy BFS, IPPO, and standard MAPPO, the proposed method achieves better scheduling safety and efficiency under the tested array setting. ? 2026 IEEE.