Ever-increasing requirements for modern-day intelligent transportation systems bring new challenges for automotive functions. Connected vehicular applications have recently become extremely important for addressing challenges in the fields of safety and efficiency on the road. However, the use of connected applications is limited due to the unreliable nature of communication links in vehicular environments. Existing channel estimation and link adaptation approaches are mainly based on utilizing the expected channel statistics or on extrapolating the previously estimated parameters. They are, however, not well suited for vehicular communications with non-stationary channel conditions, which vary significantly within channel estimation intervals due to shadowing by other vehicles, multi-path propagation and high Doppler frequency shifts. This motivates a search for new ways to improve vehicle-to-vehicle communications. The concept developed in this thesis addresses this problem by using data from the on-board perception system as an additional source of information to improve vehicleto-vehicle communications. The use of positioning data from on-board sensors has shown its potential in railway, cellular and satellite communications systems for scenarios where one dynamic communication partner moves along a known trajectory. In this thesis, the information about the surrounding environment from the perception system of a modern automated vehicle is used to predict changes in direct-link communications among highly dynamic partners. The perceived information is assumed to be exchanged between the functional components of the automated vehicle, including the communication subsystem, via standardized interfaces. This approach enables a range of improvements via sensor-aided predictive communication algorithms. In this thesis, the sensor-based predictive compensation of Doppler frequency shift and the prediction of predominantly line-of-sight or non-line-of-sight communication conditions are investigated as two illustrative classes of sensor-aided algorithms. In the first class, the perceived relative velocity of relevant road objects is used to improve compensation of the time-varying Doppler-shift in the received dominant signal components. In the second class, the information about the expected changes in the perceived road scenario is used to predict variations in the communication quality, which are associated with transitions between the line-of-sight and non-line-of-sight communication regions. A thorough investigation of the selected algorithms for different road scenarios and use-cases presents a range of benefits in the use of sensor data for highly-dynamic vehicular communications. The results obtained show that the proposed sensor-based predictive methods outperform existing reference solutions in the scenarios considered and can be beneficially applied to different communication layers, data dissemination models (single-hop, multi-hop), types of cooperative vehicular applications (platooning, cooperative collision avoidance) and driving scenarios (congested traffic, free flow; rural road or highway).