Evaluating integrity of vehicle pose estimates via semantic labels
Subject matter disclosed herein may relate to systems, devices and/or processes for monitoring and/or evaluating vehicle pose integrity for use with an autonomous driving controller. A system may include memory, the memory comprising one or more memory devices and a processor coupled to the one or more memory devices, the processor configured to access, from the one or more memory devices, sensor measurements generated by one or more sensors mounted in a vehicle. The system may additionally observe one or more semantic classes to one or more physical entities based on the generated sensor measurements and extract one or more semantically-labeled map parameters from an electronic map the system may additionally determine a correlation between the one or more observed semantic classes and one or more expected classes extracted from the semantically-labeled map parameters.
1 . A vehicle pose integrity monitor for use with an autonomous driving controller, comprising:
a memory, the memory comprising one or more memory devices; and
a processor coupled to the one or more memory devices, the processor configured to:
access, from the one or more memory devices, sensor measurements generated by one or more sensors mounted in a vehicle;
observe one or more semantic classes to one or more physical entities based on the generated sensor measurements;
extract one or more semantically-labeled map parameters from an electronic map;
determine a correlation between the one or more observed semantic classes and one or more expected classes extracted from the semantically-labeled map parameters based on a computation of true positive rate counts, false positive rate counts, true negative rate counts, and false negative rate counts;
compute a localization integrity based on the correlation; and
generate one or more operations to direct at least one autonomous driving application to perform one or more functions to influence a motion vector of the vehicle.
2 . The vehicle pose integrity monitor of claim 1 , wherein the one or more physical entities comprises:
a non-drivable region, a drivable region, a structure, a lane marking, a natural object or a traffic sign, or a combination thereof.
3 . The vehicle pose integrity monitor of claim 1 , wherein the processor is further configured to:
execute one or more operations to plan motion of the vehicle based on the computed localization integrity.
4 . The vehicle pose integrity monitor of claim 1 , wherein the processor is further configured to:
compute one or more vehicle pose hypotheses; and
compute positive or negative deviations from the computed one or more vehicle pose hypotheses.
5 . The vehicle pose integrity monitor of claim 4 , wherein the processor is further configured to:
compute at least one sample point to represent a deviation from the computed one or more vehicle pose hypotheses.
6 . The vehicle pose integrity monitor of claim 5 , wherein the processor is further configured to:
compute a correlation coefficient for each of the one or more computed vehicle pose hypotheses.
7 . The vehicle pose integrity monitor of claim 6 , wherein the one or more computed vehicle pose hypotheses comprises a plurality of computed vehicle pose hypotheses, and wherein the processor is further configured to:
compute a transformation of a distribution of the plurality of computed vehicle pose hypotheses.
8 . The vehicle pose integrity monitor of claim 7 , wherein the transformation operates to transform a nonlinear function that represents, or is derived from, the distribution of the plurality of computed vehicle pose hypotheses.
9 . The vehicle pose integrity monitor of claim 8 , wherein the computed vehicle pose hypotheses comprises a critical input for generating the one or more expected semantic classes.
10 . The vehicle pose integrity monitor of claim 9 , wherein the transformation operates to create a weighted average of each of the one or more computed vehicle pose hypotheses.
11 . A method to provide vehicle pose integrity for use with an autonomous driving controller, the method being performed by one or more processors and comprising:
accessing, from one or more memory devices, sensor measurements generated by one or more sensors mounted in a vehicle;
observing one or more semantic classes to one or more physical entities based on the generated sensor measurements;
extracting one or more semantically-labeled map parameters from an electronic map;
computing a correlation between the one or more observed semantic classes and one or more expected classes extracted from the semantically-labeled map parameters based on a computation of true positive rate counts, false positive rate counts, true negative rate counts, and false negative rate counts;
computing a localization integrity based on the correlation; and
generating one or more operations to direct at least one autonomous driving application to perform one or more functions to influence a motion vector of the vehicle.
12 . The method of claim 11 , wherein the one or more physical entities comprises: a non-drivable region, a drivable region, a structure, a lane marking, a natural object or a traffic sign, or a combination thereof.
13 . The method of claim 11 , further comprising:
executing one or more operations to plan motion of the vehicle based on the computed localization integrity.
14 . The method of claim 11 , further comprising:
computing one or more vehicle pose hypotheses; and
computing positive or negative deviations from the computed one or more vehicle pose hypotheses.
15 . The method of claim 14 , further comprising:
computing at least one sample point to represent a deviation from the computed one or more vehicle pose hypotheses.
16 . The method of claim 15 , further comprising:
computing a correlation coefficient for each of the one or more computed vehicle pose hypotheses.
17 . The method of claim 16 , wherein the one or more computed vehicle pose hypotheses comprises a plurality of computed vehicle pose hypotheses, and further comprising:
computing a transformation of a distribution of the plurality of computed vehicle pose hypotheses.
18 . The method of claim 17 , wherein the transformation operates to transform a nonlinear function that represents, or is derived from, the distribution of the plurality of computed vehicle pose hypotheses.
19 . The method of claim 18 , wherein the computed vehicle pose hypotheses comprises a critical input for generating the one or more expected semantic classes.
20 . A non-transitory computer-readable medium comprising encoded program instructions for causing one or more processors of a vehicle pose integrity monitor to perform at least the following:
access sensor measurements generated by one or more sensors mounted in a vehicle;
observe one or more semantic classes to one or more physical entities based on the generated sensor measurements;
extract one or more semantically-labeled map parameters from an electronic map;
compute a correlation between the one or more observed semantic classes and one or more expected classes extracted from the semantically-labeled map parameters based on a computation of true positive rate counts, false positive rate counts, true negative rate counts, and false negative rate counts;
compute a localization integrity based on the correlation; and
generate one or more operations to direct at least one autonomous driving application to perform one or more functions to influence a motion vector of the vehicle.