Johnathan A. Tucker

In the future I would like to investigate the real-time generation of 3D dynamic scene graphs.

Currently, simultaneous localization and mapping (SLAM) algorithms are the most common way of giving a robot an understanding of the scene that it is in. An extension of the traditional SLAM algorithm is 3D dynamic scene graphs (DSGs). In 3D DSGs nodes represent spatial concepts and the edges represent relations between these spatial concepts. For example, if a chair is adjacent to a wall in a house, then the chair and wall would be nodes with the adjacency being the edge. DSGs also take advantage of layering to create levels of abstractions. Continuing with the house example, the house might be the top layer with rooms as the layer underneath. Within this “rooms” layer, the nodes would be the rooms themselves, and the edges would describe how they are connected. This results in the ability to abstract the scene based on the layer of interest. Another important result of 3D DSGs is that they provide the robot with an understanding of semantic labels (e.g. chair) for spatial concepts. As humans, we use this semantic understanding of the world for more efficient processing. For example, if I would like to locate a set of keys, I would use the fact that the keys node is on top of the table node to streamline my search. Providing robots with this semantic understanding is imperative to human-level robot perception. 3D DSGs are among the most promising ways to provide robust human-level perception to a robot although there is a key open problem in the research. These graphs currently cannot be generated incrementally or in real-time.

---

Past Work

GPS/GNSS Interference Power Difference of Arrival (PDOA) Localization Weighted via Nearest Neighbors

The Global Positioning System (GPS) is listed by the US Department of Homeland Security as critical infrastructure. These are essential assets that underpin much of modern society and that, if removed, would have a debilitating impact on our physical health or economic system. This is no exaggeration as GPS timing synchronization is used to coordinate stock exchanges, and as the navigation capabilities it provides are used daily by nearly every smartphone user. Despite GPS being listed as critical infrastructure, it is very easy to interfere with it. Radio frequency interference (RFI) in the form of jamming or spoofing is so incredibly accessible that emitters, made to work with car adapters, capable of incapacitating vehicle tracking exist despite them being illegal. There are several methods for locating RFI emitters, and power difference of arrival (PDOA) is one of the simplest and more raw ways of achieving localization. PDOA is generally used because it requires less data to provide a solution than other alternatives, can be effective against both jamming and spoofing, and works against a variety of RFI signals, such as broadband noise or continuous wave. This paper improves on the PDOA approach by incorporating the saturation of the receivers and the nearest neighboring receivers to adjust the non-linear least squares PDOA solution. Receivers are said to be saturated when their automatic gain control has reached their power floor. This means that if a receiver is saturated it cannot provide an accurate PDOA estimate because there is an infinite amount of locations within a region close to the receiver that the emitter could be located. However, when the nearest neighbors to the saturated receiver are taken into account the saturation helps weight the estimated location. The algorithm presented in this paper is tested using data from a government sanctioned event as part of an educational agreement between CU Boulder and Edwards Air Force Base. The algorithm is tested under both broadband noise and continuous wave RFI signals and proves it can improve PDOA estimation of the emitter location by a factor of 1.5.

---