Systems and methods for providing vehicle cognition
Systems and methods for providing vehicle cognition through localization and semantic mapping are provided. Localization may involve in vehicle calculation of voxel signatures, such as by hashing weighted voxel data (S 900 , S 910 ) obtained from a machine vision system ( 110 ), and comparison of calculated signatures to cached data within a signature localization table ( 630 ) containing previously known voxel signatures and associated geospatial positions. Signature localization tables ( 630 ) may be developed by swarms of agents ( 1000 ) calculating signatures while traversing an environment and reporting calculated signatures and associated geospatial positions to a central server ( 1240 ). Once vehicles are localized, they may engage in semantic mapping. A swarm of vehicles ( 1400, 1402 ) may characterize assets encountered while traversing a local environment. Asset characterizations may be compared to known assets within the locally cached semantic map. Differences of omission and commission between observed assets and asset characterizations with the local map cache ( 1860 ) may be reported to a central server ( 1800 ). Updates to the local signature cache ( 1852 ) and/or local map cache ( 1862 ) may be transmitted from the central server ( 1800 ) back down to vehicles within the swarm ( 1840 ).
1. A method of creating a vehicle localization table used by a vehicle in determining the absolute geospatial position of the vehicle via reference to the local environment of the vehicle, the method comprising:
engaging in multiple traversals of a local environment, by one or more vehicles, each vehicle having an on board machine vision sensor and an on board computer;
observing, during each traversal by a vehicle, each of a plurality of voxels, in order to generate for each of said voxels a voxel signature and an associated voxel location hypothesis, both of which are stored locally within a localization table and utilized by the vehicle for determining the vehicle's absolute geospatial position;
aggregating localization tables from said traversals;
for each observed voxel, determining a refined voxel location hypothesis combinatorially based on voxel location hypotheses sharing a common voxel signature within the aggregated localization tables; and
updating, by one or more of said vehicles, the voxel location hypothesis within the vehicle localization table, based on the refined voxel location hypothesis.
2. The method of claim 1 , in which the step of aggregating localization tables comprises transmitting, by each vehicle, localization table content comprising voxel signatures and associated location hypotheses, from the vehicle to a central data processing computer via a wide area data network.
3. The method of claim 1 , in which the step of observing, during each traversal, each of a plurality of voxels, further comprises the substep of storing locally, within the localization tables, a time at which each voxel was observed, along with signatures extracted for that particular voxel location.
4. The method of claim 1 , in which the step of determining a refined voxel location hypothesis combinatorially comprises filtering voxel location hypotheses within the aggregated localization tables sharing a common voxel signature.
5. The method of claim 4 , in which the step of determining a refined voxel location hypothesis combinatorially further comprises averaging the filtered voxel location hypotheses.
6. The method of claim 4 , in which the step of determining a refined voxel location hypothesis combinatorially further comprises filtering outlier position estimates.
7. The method of claim 2 , further comprising the step of: receiving, by each vehicle from the central data processing computer, an updated master localization signature table.