| ™ |
Hunter seekers will use DST
In Frank Herbert's sense a "hunter seeker" is an autonomous weapon
which is instructed to find a (matching) target. I'll distinguish
between an Area Hunter Seeker, and a Target Locked Hunter Seeker.
A bad idea would be a Hunter Seeker without a constraining area in which to search for and attack its target, since the sender may become the target.The more-constrained weapon, a Target Locked Hunter Seeker, has its target specified uniquely using a perhaps low resolution image taken from a possibly great distance, but in some sense live, in the field. Such a Target Locked Hunter Seeker could also make some use of stored GIS and map information as well as target shape and subtarget (bullseye) modelling data. Working toward a target tagged in initial data it moves closer, re-images area and target, and refines both path and target information. And repeats on a loop. Like an Area Hunter Seeker it may merge limited resolution map data into its work. Acquiring higher resolution imaged data it integrates remote-acquired detail into its mission with some variant of a Simultaneous Localization and Mapping (SLAM) algorithm. Over time improving the consistency of its local perception, tracking current target movement, later transitioning from path-following to bullseye aim as the high precision of current local imaging begins to approach the high precision of target model, until the sub target bullseye of the model is imposed in a model of the perceived object, finally, the weapon at a distance of low probability of failure hits the bullseye.
The Area Hunter Seeker is a slight generalization of a Target Locked Hunter Seeker. Presumably the sender knows from field intelligence that a valid target is located within a specifiable, smallish, and safely remote region of space, perhaps specified by bounds or by a center and radius. Registered to mapped spatial regions, this system uses stored GIS and map information during travel.
In either case, AHS or TLHS, mapped expectation must be compared with observed reality (i.e. using SLAM) while on the path, whether information comes from GIS and previously collected mapping data or only from increasing resolution images as the hunter seeker approaches and visually identifies its target.
In the context of DST, SLAM is almost all spatial search, merging previous data with new by search for best-corresponding points. The general idea is this: Take some number of points from previously integrated data, find their closest points in the newly-imaged points which you can quickly do using DST in George Mou's code and algorithms; this yields a set of best point-to-point correspondences which by the consistency of reality are not bad approximations of one another. To these a least squares estimator can rapidly and deterministically calculate the best-fitting transform (shift/rotate/scale) parameters to minimize those distances as a whole. Apply the estimated transform to the newly measured data and repeat until stable to optimize an estimate of camera trajectory within the original coordinate frame. Then transform new or old into each other's coordinate frame to merge datasets. Using the new merged data, adjust travel and targeting control parameters. Repeat. This is a job in which spatial search is fundamental, unavoidable, and comprises a high percentage of total computation, typically 80%.
To conclude, a Target Locked, or even an Area Hunter Seeker is an example of a space capable system.
Hence the future of warfare, and the fates of civilizations, depend highly on the DST algorithm.