A C11 implementation of a particle filter.
Initial samples are drawn from a multivariate Gaussian distribution with user-provided population mean and covariance. Noise vectors for the prediction step, used as inputs to the process model, are drawn from a zero-mean multivariate Gaussian distribution with user-provided covariance. The process noise covariance may be a function of the action. Conversely, the measurement model may be entirely arbitrary, as libpf only uses the probability density function of the posterior.
See odometry.c for an example using the 2D velocity
motion model from Probabilistic Robotics and a single landmark measurement
model.