Wireless underground sensor networks (WUSNs) facilitate remote monitoring and control of various underground environments, which are suffering from a significant reliability problem. To address this problem and relieve the ongoing networking challenges, we propose a new concept, called the magnetic induction (MI)-assisted wireless powered underground sensor network (MI-WPUSN), which integrates the advantages of MI communication techniques with those of wireless power transfer mechanisms. MI-WPUSN offers a unique platform consisting of seven envisioned devices and four distinct communication modes to provide significant reliability potential, which is constrained by its complex and challenging data collection. To unlock the potential of MI-WPUSN, we provide a systematic research roadmap for MI-WPUSN data collection, spreading from sensor deployment to multiple channel access control and to frequency-selective routing establishment. Specifically, we first summarize the state-of-the-art research advances on sensor deployment, based on which we propose a bidirectional projection-based deployment topology that achieves high network reliability and discuss the sensor numerical determination problem resorting to weighted Voronoi diagrams. We then introduce the principles for multiple channel access control and review the existing research. Finally, after introducing kernel regression methods that can easily handle the frequency-selective property of MI-WPUSN, we develop a modified Q-learning-based routing protocol. In particular, we list and analyze the underlying challenges and related future issues in each direction.