Noise modeling using machine learning

A computer-implemented method comprising:

• collecting, using one or more virtual sensors that are operating in a simulation environment, a first set of data that represents the simulation environment, wherein the simulation environment represents virtual objects in a simulated world;
• generating, using the first set of data, a point cloud wherein the point cloud represents the simulation environment;
• collecting, using one or more real-world sensors, a second set of data that represents a real-world environment, wherein the real-world sensors are physical sensors;
• generating, using the second set of data and a neural network, a noise model that represents noise in the second set of data that prevents the second set of data from accurately representing the real-world environment;
• generating, using the noise model and the point cloud, a noisy point cloud, wherein the noisy point cloud represents or describes the real-world environment;
• passing the noisy point cloud to a discriminator model that indicates whether or not the noisy point cloud was derived from a real-world environment domain or was not derived from the real-world environment domain;
• and
• training the noise model, in response to the discriminator model indicating that the noisy point cloud was not derived from the real-world environment domain, wherein training the noise model includes modifying one or more weights of the noise model.

A non-transitory computer-readable storage medium comprising instructions which, when executed by one or more processors, cause the one or more processors to:

• collect, using one or more virtual sensors that are operating in a simulation environment, a first set of data that represents the simulation environment, wherein the simulation environment represents virtual objects in a simulated world;
• generate, using the first set of data, a point cloud wherein the point cloud represents the simulation environment;
• collect, using one or more real-world sensors, a second set of data that represents a real-world environment, wherein the real-world sensors are physical sensors;
• generate, using the second set of data and a neural network, a noise model that represents noise in the second set of data that prevents the second set of data from accurately representing the real-world environment;
• generate, using the noise model and the point cloud, a noisy point cloud, wherein the noisy point cloud represents or describes the real-world environment;
• passing the noisy point cloud to a discriminator model that indicates whether the noisy point cloud was derived from a real-world environment domain or was not derived from a real-world environment domain;
• and
• training the noise model, in response to the discriminator model indicating that the noisy point cloud was not derived from the real-world environment domain, wherein training the noise model includes modifying one or more weights of the noise model.