Vehicle Collision Risk Control Using Position Probability Distributions

The patent transforms the collision risk assessment from using simple Boolean variables to using probability distribution parameters. Specifically, it models the positions of ego vehicle and obstacles as probability distributions (e.g., Gaussian distributions) and calculates collision risk based on the overlap or distance between these distributions. This parameter transformation allows the system to account for sensor noise and trajectory uncertainties while maintaining computational efficiency through closed-form probability calculations.

The patent elevates the collision risk assessment from a one-dimensional Boolean decision (collision/no collision) to a multi-dimensional probabilistic framework. It introduces probability distributions over position, velocity, and acceleration, and computes collision risk as a continuous value based on the statistical relationship between these distributions. This dimensional expansion enables the system to capture uncertainty and partial overlap scenarios that Boolean logic cannot represent, while maintaining real-time performance through efficient probability computation.

The patent replaces complex mechanical trajectory simulation with statistical probability distribution methods. Instead of performing detailed physics-based simulations that would require significant computational resources, it models trajectories as probability distributions and uses statistical measures (such as mean, variance, and distribution overlap) to assess collision risk. This substitution maintains high prediction accuracy by capturing the essential uncertainty characteristics while dramatically reducing computational time through closed-form mathematical solutions.