Reliable Droplet Routing in Digital Microfluidic Biochips by Reinforcement Learning

In the past decade, digital microfluidic biochips (DMFBs) have emerged as a transformative technology for laboratory automation, yet their clinical adoption remains constrained by reliability challenges stemming from electrode degradation and droplet cross-contamination. In this paper, we propose a contamination-aware cooperative multi-agent reinforcement learning (CaCMARL) framework that addresses the dual challenge of real-time droplet manipulation and cross-contamination minimization. We design a problem-specific observation space and reward function for agents to accomplish the droplet routing task, while minimizing both cross-contamination and completion steps. We accelerate agent training through the introduction of a curriculum learning mechanism. Extensive simulations demonstrate that our proposed method achieves 47%-58% reduction in cross-contamination events compared to the state-of-the-art method while maintaining similar task success rates and completion steps.