Jeremy Ludwig, Stottler Henke Associates, Inc.; Bart Presnell, Stottler Henke Associates, Inc.; Richard Stottler, Stottler Henke Associates, Inc.
Keywords: space situational awareness, intent recognition, search, virtual assistant
Abstract:
The complexity of space operations has increased dramatically in the past decade. Space has become an increasingly congested and contested environment as nations and commercial entities have become more and more reliant on space-based capabilities. Enhanced space-domain awareness will be necessary to manage these challenges, but the tools and processes available today will be insufficient for the future. For example, there is an abundance of information available to space operators, but it can be underutilized given the operational tempo and the complexity of decisions being made at vast distances and orbital speeds. As part of the DARPA Hallmark program, our objective was to develop a virtual space assistant (VIRSA) to enhance decision making in space operations by (1) filtering, searching, and synthesizing heterogeneous data (2) operating both proactively in the background and on demand (3) providing tailored assistance to operators for tasks that are time-sensitive, data-intensive, recurring, or otherwise challenging, and (4) reducing operator workload with an intuitive user interface. Our requirements necessitated that the virtual assistant be tailored to real-world operational tasks, complementary to existing capabilities, intuitive, unobtrusive, quick, reliable, and convenient.
One of the focus areas of the DARPA Hallmark program is the development and evaluation of testbeds and tools for space operations concepts. The participants include two testbeds, an ontology team, a number of tool teams, and two cognitive evaluation teams. It should be noted that this abstract presents a narrow slice of the overall program; there are other participants in the Hallmark program and additional focus areas that are outside the scope of this abstract. Two testbeds form the technological foundation of Phase I. Each of the testbeds use a distinct cloud-based technology stack to deploy tools, support inter-tool communication, and provide access to shared data. Building on the testbeds, a variety of teams developed a number of tools (including VIRSA) that visualize and analyze data to support decision-making during space operations. The tools work together in a complementary fashion to provide the full range of functionality needed on a space operations floor. A series of evaluation events allowed space operations personnel to test the Hallmark environment, using the testbeds, tools, and data to make real-time decisions during situational awareness and command and control scenarios.
One issue with developing a virtual assistant for space operations is that the functionality requested by operators is limited only by the imagination. It is also difficult to know ahead of time what tasks are going to be time-sensitive, data-intensive, recurring, or otherwise challenging in the Hallmark environment until after the tools and scenarios are developed. Two features became the focus of the Phase I development and demonstration efforts: RSO Summary and Keyword Search. RSO Summary is an at-a-glance summary of key information about a resident space object. Operators used this information to quickly assess a situation, solve problems in the face of unexpected events, dynamically re-plan, and anticipate implications of potential courses of action. Keyword Search addressed the need to perform efficient searches across tools and data sources.
A series of evaluation events held every three months during the Hallmark program, which allowed space operations personnel to test the Hallmark environment, using the testbeds, tools, and data to make real-time decisions during situational awareness and command and control scenarios. The evaluation events were significant undertakings, where scenarios were presented in the context of a simulated operations floor and staffed with personnel with real-world experience. Their expertise was matched against increasingly complex scenarios and an increasingly capable Hallmark environment. The feedback for VIRSA from the Phase I events confirmed the utility of the RSO Summary and Keyword Search results along with a list of requests for Phase II.
The VIRSA Phase II design included adding three new features as suggested by the Phase I results, in additional to the continued refinement of RSO Summary and Keyword Search. The first new feature is Intent Recognition. When a user types text into the search bar, VIRSA should try to answer the question or perform the task for a narrow, pre-defined set of use case. If that fails, VIRSA falls back on performing search. The second new feature is a Topic Service, which tracks popular topics from sources such as search and chat and displays them as autocomplete options when operators type in the search box. The idea is to provide visibility and ease of use for trending topics. The third new feature is Search Templates, which preemptively suggests intent-based searches to the operator as additional autocomplete options. Taken together, these three features move the functionality of VIRSA closer to that of familiar general-purpose search engines like Google.
VIRSA illustrates concrete features that a virtual assistant would use to improve decision-making on the space operations floor. Within the context of the Hallmark environment, we have identified the highest impact initial features, evaluated these features in a series of week-long evaluation events, identified additional assistant functionality for Phase II, and then developed and evaluated this functionality while continuing to refine the Phase I implementation. Based on the realism and complexity of the evaluation events, we believe that the implemented features will be useful in other space operations environments as well. However, we do see several challenges in transitioning an assistant like VIRSA into operational use.
Date of Conference: September 14-17, 2021
Track: SSA/SDA