Liana Suantak

An optimal motion planning method for computer-assisted surgical training

Graphical abstractDisplay Omitted HighlightsAn optimal motion planning method for computer-assisted surgical training is developed and validated in the paper.The method generates shortest, collision-free trajectories for laparoscopic instrument movements in the rigid block world used for hand-eye coordination tasks.Optimal trajectories are displayed on a monitor to provide continuous visual guidance for optimal navigation of instruments. This paper focuses on the development and validation of an optimal motion planning method for computer-assisted surgical training. The context of this work is the development of new-generation systems that combine artificial intelligence and computer vision techniques in order to adjust the learning process to specific needs of a trainee, while preventing a trainee from the memorization of particular task settings. The problem described in the paper is the generation of shortest, collision-free trajectories for laparoscopic instrument movements in the rigid block world used for hand-eye coordination tasks. Optimal trajectories are displayed on a monitor to provide continuous visual guidance for optimal navigation of instruments. The key result of the work is a framework for the transition from surgical training systems in which users are dependent on predefined task settings and lack guidance for optimal navigation of laparoscopic instruments, to the so called intelligent systems that can potentially deliver the utmost flexibility to the learning process. A preliminary empirical evaluation of the developed optimal motion planning method has demonstrated the increase of total scores measured by total time taken to complete the task, and the instrument movement economy ratio. Experimentation with different task settings and the technical enhancement of the visual guidance are subjects of future research.

A coevolutionary approach to course of action generation and visualization in multi-sided conflicts

The current state of military operations includes many stability and support (SASO), multi-sided conflicts. The research presented in this paper attempts to address this complex environment by creating a SASO simulation, coevolutionary generation of courses-of-actions (COAs) for each side, and visualization tools for analysis of the resulting COAs. The SASO simulation is significantly different from previous systems because it incorporates non-conventional warfare units such as terrorists and media. The coevolution algorithm is different because it allows all sides of the conflict to evolve their COAs. The visualization tools are important because SASO doctrine is not as well developed as conventional warfare doctrine. Therefore, visual analysis and understanding of a system that is not well defined provides insight for future modeling and verification.

Modeling and simulation of stability and support operations (SASO)

Stability and support operations (SASO) are becoming increasingly important in modern military operations. Conflicts are no longer comprised solely of two opposing sides engaged in combat on an open battlefield. Instead, they are more likely to involve groups sharing various alliances and relationships each pursuing a range of different goals. The Sheherazade SASO wargaming engine presented here: a) incorporates subjective criteria for scoring course of action (COA) success such as the animosity between factions and attitudes of locales, b) uses nontraditional units such as refugees, media and information operators, and c) employs a coevolutionary genetic algorithm in modeling the dynamics of the complex multisided simulation for generating COAs. This paper outlines our approach towards the development of a wargaming model that handles the more complex and computationally demanding arena of SASO.

Decision Support Using Deterministic Equivalents of Probabilistic Game Trees

We have developed a game-theory driven decision-support tool that builds probabilistic game trees automatically from user-defined actions, rules, and states. The result of evaluating the paths in the game tree is a series of decisions which forms a decision-path representing an epsilon-Nash-Equilibrium. The algorithm uses certainty-equivalents to handle trade-offs between expected rewards and risks, effectively modeling the probabilistic game tree as deterministic. The resulting decision-paths correspond to player actions in the scenario. These sets of actions can be used as search patterns against a real-world database. A match to one of these patterns indicates an instance of novel behavior patterns generated by the game-theory driven decision support tool. This particular paradigm could be applied in any domain that requires anticipating and responding to adversarial agents with uncertainty, from mission planning to emergency responders to systems configuration.

On the Extraction and Analysis of a Social Network with Partial Organizational Observation

The behavior of an organization may be inferred based on the behavior of its members, their contacts, and their connectivity. One approach to organizational analysis is the construction and interpretation of a social network graph, where entities of an organization (persons, vehicles, locations, events, etc.) are nodes, and edges represent varying kinds of connectivity between entities. This paper describes a transformation based approach to the extraction of a social network graph, where the original data comprising (partial) observation of the organization are embedded on a graph with a different ontology, and with many entities and edges that are unrelated to the organization of interest. Social network extraction allows the inference of implied relationships, and the selection of relationships relevant for intended analysis techniques. The analysis of the resulting social network graph is based on organizational and individual analysis, in order to permit an advanced user to draw conclusions regarding the behavior of the organization, based on established social network graph metrics. The results of the paper include a discussion of the complexity of analysis, and how the observation data graph is pruned in order to scale the application of analysis algorithms.

Visualization Tools to Adapt to Complex Military Environments

This chapter addresses adaptation to dynamic, novel and uncertain military environments. These environments require a shift from a maneuver warfare paradigm to an asymmetric world where shifting alliances, questionable civilian loyalties, opaque cultures, and the requirement to maintain peace one day and combat the next makes for a particularly confusing situation. This new warfare paradigm requires adaptation to an uncertain, complex environment. The initial section discusses a general cognitive model of visualization called RAVENS and its importance for adaptation developed specifically to address complex military environments. RAVENS posits that humans are inherently flexible decision makers and situation awareness depends on the ability of humans to create narrative visualizations that capture the overall context of complex military environments. Using the framework as a guideline, we will examine two important visualization research programs whose purpose is to allow military operators to rapidly adapt to volatile situations. The first program investigates cognitive effects such as the framing bias and their possible interactions with a variety of display concepts during a series of missile defense simulations. The experimenters presented risk as a spatial representation of uncertainty and target value that emphasized either expected population lost or expected population saved. The second program investigated the feasibility of using visualizations generated from Scheherazade (a coevolutionary algorithm) to aid MI analysts in predicting emergent tactics of terrorist groups during urban operations. Finally, we discuss the value of these approaches for providing coherent narrative understanding as called for in the RAVENS model.

A hybrid architecture for visualization and decision making in battlespace environments

This article presents a hybrid software/hardware architecture for commanders decision support in tactical operations. The architecture builds on the symbolic, object-oriented visualization software called Advanced Tactical Architecture for Combat Knowledge System (ATACKS). The extension discussed here is the design of a real-time robot agent layer that interacts wirelessly with ATACKS. This layer enacts decisions made by software agents (wargamers), continuously relays the execution states back to ATACKS, and updates its actions as advocated by replanning algorithms. The software layer is briefly described followed by the specification of the real-time requirements for the robotic architecture. The design and implementation are given with a small example that illustrates the hybrid systems operation.

Queral Networks: Toward an Approach for Engineering Large Artificial Neural Networks

A generalization of an artificial neuron is introduced in this paper. Called the queron, this abstraction is the basic computational node of Queral Networks (QN). QNs are introduced as a parallel architecture expected to be an improvement upon Artificial Neural Networks (ANN). The fundamental properties of QNs are presented here: reusability, complexity management and human-readability. It is expected that this proposed architecture will allow the engineering of large, highly parallel computer systems with the computational benefits of ANNs while overcoming the challenge of developing ANNs. A brief case study is given to illustrate the QN concept.