This report is a part of the project called PRECOG with Charles River Analytics. Probabilistic programming languages provide a rich modeling framework for defining and solving sequential decision-making problems under uncertainty. Figaro can define a decision model including continuous and discrete random variables and allow composition of hybrid mixtures of continuous and discrete random variables, offering flexible and realistic decision models that reflect complex real-world scenarios. However, such rich and powerful modeling capability brings significant technical challenges in design and implementation of algorithms for solving sequential decision problems. This report defines the semantics of sequential decision problems in terms of the graphical model framework called influence diagrams, clarifies some of the technical challenges imposed by allowing arbitrary composition of continuous and discrete random variables in the model, a new approach called model sampling with online planning, and gives and initial assessment of the proposed methods. In addition, the possible future work and directions to extend the current approach are given in the conclusion. The final outcomes accompanied with this report are a set of python scripts that prototype the proposed method (model sampling and online planning for hybrid influence diagrams), and our approximate discrete model inference algorithm for solving large scale influence diagrams (join graph decomposition bounds for influence diagrams), which are also described at the end of this report.