Gilbert L. Peterson’s research while affiliated with U.S. Air Force Institute of Technology and other places

Proper process parameter calibration is critical to the success of fused deposition modeling (FDM) three‐dimensional (3D) printing, but is time‐consuming and requires expertise. While existing systems for autonomous calibration have demonstrated success in calibrating for a single objective, users may need to balance multiple conflicting objectives. Herein, an easily deployable, camera‐based system for autonomous calibration of FDM printers that optimizes for both part quality and completion time is presented. Autonomous calibration is achieved through a novel, multifaceted computer vision characterization and a multitask learning extension to Bayesian optimization. The system is demonstrated on four popular filament types using two distinct 3D printers. The results show that the system significantly outperforms manufacturer calibration across the machine and material configurations, achieving an average improvement of 32.2% in quality and a 31.2% decrease in completion time with respect to a popular benchmark.

E-learning courses often suffer from high dropout rates and low student satisfaction. One way to address this issue is to use Personalized Learning Paths (PLPs), which are sequences of learning materials that meet the individual needs of students. However, creating PLPs is difficult and often involves combining knowledge graphs, student profiles, and learning materials. Researchers typically assume that the problem of creating PLPs belong to the NP-Hard class of computational problems. However, previous research in this field has neither defined the different variations of the PLP problem nor formally established their computational complexity. Without clear definitions of the PLP variations, researchers risk making invalid comparisons and conclusions when they use different metaheuristics for different PLP problems. In order to unify this conversation, this paper formally proves the NP-completeness of two common PLP variations and their generalizations and uses them to categorize recent research in the PLP field. It then presents an instance of the PLP problem using real-world data and shows how this instance can be cast into two different NP-complete variations. This paper then presents three AI strategies, solving one of the PLP variations with back-tracking and branch and bound heuristics and also converting the PLP variation instance to XCSP ³ , an intermediate constraint satisfaction language to be resolved with a general Constraint Optimization solver. This paper solves the other PLP variation instance using a greedy search heuristic. The paper finishes by comparing the results of the two different PLP variations.

Humans play games like Diplomacy for more reasons than just trying to win. For example, when they know they are losing, they may choose to aid an opponent rather than concede. This paper presents Lyre, an agent skilled at playing/communicating in Diplomacy and adept at working towards the benefit or harm of a given opponent - kingmaking. We parallel Lyre’s kingmaking with real human Diplomacy games and analyze how Lyre can best play kingmaker. Results show statistically significant benefit in Lyre’s boardplay, negotiation, and kingmaking.

Fused deposition modeling (FDM) is one of the most popular additive manufacturing (AM) technologies for reasons including its low cost and versatility. However, like many AM technologies, the FDM process is sensitive to changes in the feedstock material. Utilizing a new feedstock requires a time-consuming trial-and-error process to identify optimal settings for a large number of process parameters. The experience required to efficiently calibrate a printer to a new feedstock acts as a barrier to entry. To enable greater accessibility to non-expert users, this paper presents the first system for autonomous calibration of low-cost FDM 3D printers that demonstrates optimizing process parameters for printing complex 3D models with submillimeter dimensional accuracy. Autonomous calibration is achieved by combining a computer vision-based quality analysis with a single-solution metaheuristic to efficiently search the parameter space. The system requires only a consumer-grade camera and computer capable of running modern 3D printing software and uses a calibration budget of just 30 g of filament (~ $1 USD). The results show that for several popular thermoplastic filaments, the system can autonomously calibrate a 3D printer to print complex 3D models with an average deviation in dimensional accuracy of 0.047 mm, which is more accurate than the 3D printer’s published tolerance of 0.1–0.4 mm.

Wargames often include fog of war, i.e. hidden information, and multi-action turns, where each turn requires making multiple, sequential action choices. These properties provide unique challenges for Artificial Intelligence agents. Extensions to Monte-Carlo Tree Search (MCTS) allow it to perform well in games with hidden information as well as multi-action games. However, these extensions do not specifically consider both properties simultaneously nor how information-gaining actions could improve agent performance. Information-gaining actions are important in multi-action turns where initial actions can reveal state information, thus improving later action decisions. This paper presents enhancements to MCTS that add an information gain incentive and a risk determinization to balance locating opponent pieces while minimizing exposure to enemy fire. The information gain incentive and risk functions are implemented in Perfect Information-MCTS (PIMCTS) and Information Set-MCTS (ISMCTS) and evaluated on the multi-action hidden information game TUBSTAP. Results show that these additions improve performance over the baseline algorithms and against a Cheating MCTS implementation. KeywordsMulti-action turn-based gamesHidden InformationMonte-Carlo Tree Search

Teams of human operators and artificial intelligent agents (AIAs) in multi-agent systems present a unique set of challenges to team coordination. This research endeavors to employ a machine learning framework to estimate a set of ranks among quality goals, where the quality goals are designed to help communicate important elements of operator intent to aid the development of a Shared Mental Model among members in a multi-agent team. Using a representation referred to as the Operationalized Intent model to capture quality goals relevant to “how” the operator would like to execute the team’s mission, this paper details the development and evaluation of a random forest algorithm to estimate operator priorities. Estimation is structured as a label ranking problem in which quality goals, which constrain “how” work is to be conducted, are ranked according to their priority. Modifying an existing label ranking algorithm, we demonstrate that the Operationalized Intent Estimator-Random Forest (OIE-RF) can estimate quality goal rankings more accurately than a situation baseline which is derived by observing the variability among operators. OIE-RF demonstrates stability in dynamic testing and the ability to use explicit communication and operator identity to increase accuracy. This exploratory research opens a new avenue for improving coordination and performance of human-agent teams.

A shared mental model (SMM) is a foundational structure in high performing, task-oriented teams and aid humans in determining their teammate's goals and intentions. Higher levels of mental alignment between teammates can reduce the direct dialogue required for team success. For decision-making teams, a transactive memory system (TMS) offers team members a map of specialized knowledge, indicating source of knowledge and the source's credibility. SMM and TMS formulations aid human-agent team performance in their intended team types. However, neither improve team performance with a project team--one that requires both behavioral and knowledge integration. We present a hybrid cognitive model (HCM) for machine agents that subsumes the integrated portions of a team's transactive memory in an SMM. The unified structure of the HCM enables contextual switches during execution for machine agents, over the two cognitive formulations with comparable computational complexity of a single cognitive model. Results in a multi-agent project environment demonstrates how the HCM provides machine agents with a generalizable cognitive structure that is able to maintain fully factored belief states with minimal inter-agent communication.

Penetration Testing (pentesting) is the process of using tactics and techniques to penetrate computer systems and networks to expose any issues in their cybersecurity \cite{rsa}. It is currently a manual process requiring significant experience and time that are in limited supply. One way to supplement the shortage is through automation. This paper presents the Automated Network Discovery and Exploitation System (ANDES) which demonstrates that it is feasible to automate the pentesting process. The uniqueness of ANDES is the use of Bayesian decision networks to represent the pentesting domain and subject matter expert knowledge. ANDES conducts multiple execution cycles, which build upon previous action results. This process simulates the iterative thinking process of human attackers. Cycles begin by modeling the current belief state using Bayesian decision networks. ANDES uses these networks to select and execute an expected best action. Observed results are used to update the systems current belief state before the next cycle begins. ANDES was tested in a live-execution event, taking place within a virtual network environment designed to mimic a small business’s internal network. ANDES successfully performed a series of information gathering and remote exploit actions, across multiple network hosts, to gain access to the objective target.