|
This paper introduces a new concept that applies computational imaging techniques to laser radar for robotic perception. We observe that nearly all contemporary laser radars for robotic (i.e., autonomous) applications use pixel basis scanning where there is a one-to-one correspondence between world coordinates and the measurements directly produced by the instrument. In such systems this is accomplished through beam scanning and/or the imaging properties of focal-plane optics. While these pixel-basis measurements yield point clouds suitable for straightforward human interpretation, the purpose of robotic perception is the extraction of meaningful features from a scene, making human interpretability and its attendant constraints mostly unnecessary. The imposing size, weight, power and cost of contemporary systems is problematic, and relief from factors that increase these metrics is important to the practicality of robotic systems. We present a system concept free from pixel basis sampling constraints that promotes efficient and adaptable sensing modes. The cornerstone of our approach is agile and arbitrary beam formation that, when combined with a generalized mathematical framework for imaging, is suited to the particular challenges and opportunities of robotic perception systems. Our hardware concept looks toward future systems with optical device technology closely resembling modern electronically-scanned-array radar that may be years away from practicality. We present the design concept and results from a prototype system constructed and tested in a laboratory environment using a combination of developed hardware and surrogate devices for beam formation. The technological status and prognosis for key components in the system is discussed.
|
