In many real-world robotic scenarios, exact knowledge about a robot’s state cannot be assumed due to unmodeled dynamics or noisy sensors. Planning in belief space provides an approach that addresses this problem by tightly coupling perception and planning modules to obtain a trajectory that takes into account the stochasticity of the environment. However, existing methods are often limited to simple tasks such as the classic reach-avoid problem and are not capable of solving problems under complex spatio-temporal specifications. We address this problem of motion planning in belief space under temporal logic specifications. We present our approach on using the quantitative semantics of Risk Signal Temporal Logic (RiSTL) to generate motion plans in Gaussian belief spaces based on a Model Predictive Control (MPC) scheme. We propose a novel formulation for the risk of being inside or outside of convex polygons that allows us to specify a wide variety of predicate functions such as risk-aware reach objectives or obstacle avoidance constraints.