Facilitated by multiple sensors on mobile devices, mobile crowd sensing (MCS) relies on the mobility and characteristics of mobile users to perceive the physical world in real time, inspiring massive innovative services. Despite its prevailing deployment and great potential, conventional MCS devices transmit all sensing data to the requestors, overburdening them with high communication and computation resource consumption. This becomes even worse in practice since redundant workers are recruited for quality consideration, thus offsetting the major advantage of economical monitoring and frustrating resource-constrained requestors. This project seeks to integrate sensing and learning for MCS without the consumption of excessive resources. Specifically, given that sensing data is being collected through dispersed edge servers, blockchain-based federated learning (FL) is introduced to protect data privacy and achieve distributed machine learning (ML) with performance enhancement of trustworthiness and efficiency. The technical contributions of this research include the extension of trust from on-chain to off-chain procedures via incentive mechanism designs for eliciting trustworthy submissions from distributed edge learners. It also aims to establish instantly reliable computing environments in an off-chain manner for guaranteed efficiency of distributed ML in MCS, with both the intra-environment consensus protocol design and inter-environment interaction analysis.