Shared Vehicle Control Using Cognitive Priors for Driver Intent

The cognitive system is segmented into multiple specialized modules: world encoder for environmental perception, driver state encoder for human decision-making analysis, prediction models for specific decision parameters (latency, preferences, intent), and control signal generator. Each module handles a specific aspect of the complex prediction task, improving accuracy while managing system complexity through functional decomposition.

The patent introduces a latent representation as an intermediary that bridges sensor data and prediction outputs. This latent space encodes abstracted features of the driving environment and driver state, serving as a mediator that transforms raw sensor inputs into meaningful predictions of human decision-making characteristics without requiring direct complex modeling of all variables.

The system performs preliminary encoding of sensor data into latent representations before detailed prediction analysis. The world encoder and driver state encoder pre-process environmental and human factors into compressed latent features, reducing the computational burden on subsequent prediction models while preserving essential information for accurate human decision-making prediction.

The patent implements selective prediction of human decision-making aspects based on driving context. The system predicts only the most relevant decision parameters (such as latency, preferences, or intent) depending on the specific driving situation, rather than continuously computing all possible human decision factors, thereby optimizing computational resource usage while maintaining prediction accuracy where needed.