Q-learning based Droplet Routing in Digital Microfluidic Biochip

Digital microfluidic biochip (DMFB) is popular as laboratory on chip device. Droplet routing in DMFB is a major research area under computer aided design. Major goal of this CAD based research is to efficiently manage droplet movement on DMFB to complete the task in less time using a smaller number of unique electrodes. Application of machine learning methods in this domain is very less as traditional supervised learning mechanisms are not suitable due to lack of proper labelled training data. This article proposes a Q-learning based approach to accomplish the goal of droplet routing. DMFB board serves as environment, droplets as agent and their rewards are based on target directed movement and fluidic constraints. Initial routing trace records an accumulated rewards which later helps in identifying the right paths with optimal solution. Experimental analysis shows some encouraging results in terms of routing completion time and total unique electrode usage over the existing research reports.