Method, apparatus, and system for generating synthetic ground truth drive and sensor observation data

Patent metadata

Field	Value
Publication number	US-12498251-B2
Application number	US-202318133343-A
Country	US
Kind code	B2
Filing date	Apr 11, 2023
Priority date	Apr 11, 2023
Publication date	Dec 16, 2025
Grant date	Dec 16, 2025

Abstract

An approach is provided for generating synthetic ground truth drive and sensor observation data. The approach, for instance, receiving, by a processor, a first input specifying ground truth data indicating one or more ground truth locations of one or more map features. The approach also involves receiving a second input specifying one or more simulation characteristics. The approach further involves generating simulated drive data based on the ground truth data and the one or more simulation characteristics. For example, the simulated drive data includes (a) one or more simulated drive paths within a region of interest encompassing the one or more ground truth locations and (b) one or more simulated sensor observations of the one or more map features. The approach further involves providing the simulated drive data as an output.

First claim

What is claimed is:

1 . A method comprising:
receiving, by a processor, a first input specifying ground truth data indicating one or more ground truth locations of one or more map features; receiving a second input specifying one or more simulation characteristics; generating simulated drive data based on the ground truth data and the one or more simulation characteristics, wherein the simulated drive data includes (a) one or more simulated drive paths within a region of interest encompassing the one or more ground truth locations and (b) one or more simulated sensor observations of the one or more map features; wherein the simulated drive data is generated using a generative adversarial network (GAN); and wherein the GAN is trained on real-world driving data for a generator network of the GAN to learn to generate synthetic drive paths that a discriminator network of the GAN cannot distinguish from real-world drive paths; providing the simulated drive data to an automated map creation system to generate an estimated map; and comparing the estimated map to the ground truth data to determine a performance level of the automated map creation system.

2 . The method of claim 1 , further comprising:
improving a training of one or more feature detectors of the automated map creation system based on the performance level.

3 . The method of claim 1 , wherein the ground truth data further includes one or more attributes of the one or more map features.

4 . The method of claim 3 , wherein the one or more attributes include dimensions, a color, a type, or a combination thereof.

5 . The method of claim 1 , wherein the simulated drive path is generated based on one or more drivable surfaces determined from map data of the region of interest.

6 . The method of claim 1 , wherein the simulated drive path comprises a plurality of timestamped vehicle poses sampled at designated rate over an automatically determined trajectory over a road network in the region of interest.

7 . The method of claim 1 , wherein the one or more simulation characteristics include one or more statistical characteristics based on a positioning accuracy, a feature detection accuracy, or a combination thereof.

8 . The method of claim 1 , wherein the one or more simulation characteristics include a number of turns, a vehicle speed, a traffic density, a weather parameter, or a combination thereof on the one or more simulated drive paths.

9 . The method of claim 1 , wherein at least one simulated drive path of the one or more simulated drive paths is labeled as a true drive path based on a road network in the area of interest.

10 . The method of claim 1 , wherein the one or more simulated sensor observations are based on odometry, and wherein the odometry indicates a relative vehicle pose displacement from one timestamp to a next timestamp.

11 . The method of claim 1 , wherein the one or more simulated sensor observations are based on an absolute vehicle position determined at a timestamp.

12 . The method of claim 1 , wherein the one or more simulated observations includes a location measurement of the one or more map features, a measurement of one or more attributes of the one or more map features, or a combination thereof.

13 . The method of claim 1 , wherein the one or more simulation characteristics include a false positive rate, a false negative rate, or a combination thereof; and wherein the one or more simulated sensor observations are generated based on the false positive rate, the false negative rate, or a combination thereof.

14 . The method of claim 1 , further comprising:
applying at least one random perturbation to the one or more simulated drive paths, the one or more simulated sensor observations, or a combination thereof.

15 . An apparatus comprising:
at least one processor; and at least one memory including computer program code for one or more programs, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform at least the following,
receive a first input specifying ground truth data indicating one or more ground truth locations of one or more map features;
receive a second input specifying one or more simulation characteristics;
generate simulated drive data based on the ground truth data and the one or more simulation characteristics, wherein the simulated drive data includes (a) one or more simulated drive paths within a region of interest encompassing the one or more ground truth locations and (b) one or more simulated sensor observations of the one or more map features; wherein the simulated drive data is generated using a generative adversarial network (GAN); and wherein the GAN is trained on real-world driving data for a generator network of the GAN to learn to generate synthetic drive paths that a discriminator network of the GAN cannot distinguish from real-world drive paths; and
provide the simulated drive data to an automated map creation system to generate an estimated map; and
compare the estimated map to the ground truth data to determine a performance level of the automated map creation system.

16 . The apparatus of claim 15 , wherein the apparatus if further caused to:
improve a training of one or more feature detectors of the automated map creation system based on the performance level.

17 . The apparatus of claim 15 , wherein the ground truth data further includes one or more attributes of the one or more map features.

18 . A non-transitory computer-readable storage medium, carrying one or more sequences of one or more instructions which, when executed by one or more processors, cause an apparatus to perform:
receiving a first input specifying ground truth data indicating one or more ground truth locations of one or more map features; receiving a second input specifying one or more simulation characteristics; generating simulated drive data based on the ground truth data and the one or more simulation characteristics, wherein the simulated drive data includes (a) one or more simulated drive paths within a region of interest encompassing the one or more ground truth locations and (b) one or more simulated sensor observations of the one or more map features; wherein the simulated drive data is generated using a generative adversarial network (GAN); and wherein the GAN is trained on real-world driving data for a generator network of the GAN to learn to generate synthetic drive paths that a discriminator network of the GAN cannot distinguish from real-world drive paths; providing the simulated drive data to an automated map creation system to generate an estimated map; and comparing the estimated map to the ground truth data to determine a performance level of the automated map creation system.

19 . The non-transitory computer-readable storage medium of claim 18 , wherein the apparatus is caused to further perform:
improving a training of one or more feature detectors of the automated map creation system based on the performance level.

20 . The non-transitory computer-readable storage medium of claim 18 , wherein the ground truth data further includes one or more attributes of the one or more map features.

Assignees

• Here Global Bv

Inventors

• Huffman Landis
• Kadam Nilesh
• Williamson Dennis Scott
• Raggi Roberto
• Lynch James D
• Kavuri Prasad
• Saxena Shishir B
• Boddhu Sanjay Kumar

Classifications

• G06F2111/08

  Probabilistic or stochastic CAD · CPC title

• G06F30/27

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G01C21/3844 Primary

  Data obtained from position sensors only, e.g. from inertial navigation · CPC title

• G01C21/3804 Primary

  Creation or updating of map data · CPC title