Using a GAN discriminator for anomaly detection
Systems, methods, and other embodiments described herein relate to improving the accuracy of sensor data by using a discriminator to identify the presence of anomalies. In one embodiment, a method includes acquiring sensor data about a roadway. The method includes determining a presence of an anomaly within the sensor data using a discriminator from a generative adversarial network (GAN). The method includes generating an indicator for the sensor data according to whether the anomaly is present. The method includes providing the indicator with the sensor data.
1 . A detection system for identifying anomalies, comprising:
one or more processors;
a memory communicably coupled to the one or more processors and storing instructions that, when executed by the one or more processors, cause the one or more processors to:
acquire sensor data about a roadway;
determine a presence of an anomaly within the sensor data using a discriminator from a generative adversarial network (GAN), wherein the anomaly is an aberration within the sensor data associated with an error in perceiving the roadway;
generate an indicator for the sensor data according to whether the anomaly is present; and
provide the indicator with the sensor data.
2 . The detection system of claim 1 ,
wherein the sensor data includes one or more of orthographic images of the roadway, trace data comprised of frames specifying locations and poses of one or more probe vehicles along the roadway, and detections at the frames that are identified aspects of the roadway, including lane boundaries, road boundaries, and road markings.
3 . The detection system of claim 1 , wherein the instructions to determine the presence of the anomaly include instructions to process the sensor data using the discriminator to distinguish when the sensor data includes the anomaly by leveraging learned awareness of the discriminator in identifying between real information and synthetic information, and
wherein the instructions to generate the indicator include instructions to flag the sensor data to identify a presence of the anomaly.
4 . The detection system of claim 1 , wherein the discriminator is a machine-learning model that classifies the sensor data according to a presence of the anomaly.
5 . The detection system of claim 1 , wherein the instructions further include instructions to train, prior to determining the presence of the anomaly, a probe model of the GAN to synthesize probe data and the discriminator to distinguish between real probe data and synthesized probe data.
6 . The detection system of claim 5 , wherein the instructions to train the probe model include instructions to use an output of the discriminator to adapt the probe model to improve synthesizing the probe data.
7 . The detection system of claim 1 , wherein the instructions to provide the indicator with the sensor data include instructions to perform one or more of communicating a request to a set of probe vehicles to acquire additional data about the roadway in an area associated with the anomaly, and generating a map using accumulated data that has been identified as not including the anomaly.
8 . The detection system of claim 7 , wherein the instructions to provide the indicator include instructions to use the map to control a vehicle to navigate the roadway.
9 . A non-transitory computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to:
acquire sensor data about a roadway;
determine a presence of an anomaly within the sensor data using a discriminator from a generative adversarial network (GAN), wherein the anomaly is an aberration within the sensor data associated with an error in perceiving the roadway;
generate an indicator for the sensor data according to whether the anomaly is present; and
provide the indicator with the sensor data.
10 . The non-transitory computer-readable medium of claim 9 ,
wherein the sensor data includes one or more of orthographic images of the roadway, trace data comprised of frames specifying locations and poses of one or more probe vehicles along the roadway, and detections at the frames that are identified aspects of the roadway, including lane boundaries, road boundaries, and road markings.
11 . The non-transitory computer-readable medium of claim 9 , wherein the instructions to determine the presence of the anomaly include instructions to process the sensor data using the discriminator to distinguish when the sensor data includes the anomaly by leveraging learned awareness of the discriminator in identifying between real information and synthetic information, and
wherein the instructions to generate the indicator include instructions to flag the sensor data to identify a presence of the anomaly.
12 . The non-transitory computer-readable medium of claim 9 , wherein the discriminator is a machine-learning model that classifies the sensor data according to a presence of the anomaly.
13 . The non-transitory computer-readable medium of claim 9 , wherein the instructions further include instructions to train, prior to determining the presence of the anomaly, a probe model of the GAN to synthesize probe data and the discriminator to distinguish between real probe data and synthesized probe data.
14 . A method, comprising:
acquiring sensor data about a roadway;
determining a presence of an anomaly within the sensor data using a discriminator from a generative adversarial network (GAN), wherein the anomaly is an aberration within the sensor data associated with an error in perceiving the roadway;
generating an indicator for the sensor data according to whether the anomaly is present; and
providing the indicator with the sensor data.
15 . The method of claim 14 ,
wherein the sensor data includes one or more of orthographic images of the roadway, trace data comprised of frames specifying locations and poses of one or more probe vehicles along the roadway, and detections at the frames that are identified aspects of the roadway, including lane boundaries, road boundaries, and road markings.
16 . The method of claim 14 , wherein determining the presence of the anomaly includes processing the sensor data using the discriminator to distinguish when the sensor data includes the anomaly by leveraging learned awareness of the discriminator in identifying between real information and synthetic information, and
wherein generating the indicator includes flagging the sensor data to identify a presence of the anomaly.
17 . The method of claim 14 , wherein the discriminator is a machine-learning model that classifies the sensor data according to a presence of the anomaly.
18 . The method of claim 14 , further comprising:
training, prior to determining the presence of the anomaly, a probe model of the GAN to synthesize probe data and the discriminator to distinguish between real probe data and synthesized probe data.
19 . The method of claim 18 , wherein training the probe model includes using an output of the discriminator to adapt the probe model to improve synthesizing the probe data.
20 . The method of claim 14 , wherein providing the indicator with the sensor data includes one or more of communicating a request to a set of probe vehicles to acquire additional data about the roadway in an area associated with the anomaly, and generating a map using accumulated data that has been identified as not including the anomaly.