Nicholas Gans

In the spring of 2020, when campus was closed due to Covid-19, a walk through a university building and one of its mechanical engineering labs would have taken us through deserted and eerie spaces....

The U.S. Army improved upon modeling with the Tank Accuracy Model (TAM) in 1993 to calculate the probability of a stationary tank firing at a stationary target. Within TAM, Lay Error (LE) represents a gunner's inability to lay a reticle exactly on a target's center and is a critical factor in determining a hit. Traditional TAM LE models are still widely applied but are now seen as insufficient. These models fail because the original TAM system was either incorrect or performs poorly, having been designed to operate only under a narrow set of firing conditions. This design limitation means the system is not built to operate effectively outside those specific parameters, nor was it ever originally designed to account for Artificial Intelligence (AI) or its influence on modern targeting technologies. The proposed work introduces an Unreal5-based simulation and an associated automated target recognition (ATR) dataset, featuring pixel masks of targets under varying ranges, zoom levels, occlusions, and fog levels. This paper designs LE-task-specific ATR models using object detection backbones (YOLOV4, YOLOX-Tiny, and YOLOX) trained on the new ATR data. Experiments evaluate traditional TAM and optimizable model LE estimation against human and AI aiming data, compiling the effect of changing contrast, target occlusion, and target types. Results show that newer LE-task-specific AI models (YOLOX and YOLOX-Tiny) achieve low LE across varying ranges and demonstrate resilience to changing firing conditions, outperforming both YOLOV4 and human participants. Optimized versions of LE models also provide superior LE estimation compared to previous TAM models, reducing error rates by over 80% for the original TAM use case.

This study pursued two key objectives: first, to create an advanced model that accurately represents a human’s precision in aligning sight crosshairs with the center of a target, otherwise known as lay error; second, to investigate how various engagement conditions (such as target shape, size, range, and motion) influence lay error and determine whether a single error model could be reliably applied across different targets. To address these objectives, a photo-realistic simulation environment using Unreal Engine was developed featuring four different targets, four motion configurations, four levels of zoom, four ranges, three levels of contrast, and four levels of obstructions. After prototyping and evaluation, the simulation environment was used to collect lay error metrics as 110 IRB-approved subjects aligned crosshairs on targets under various conditions. Subjects fired a total of 11,088 shots throughout a set of three different data collection sessions that were completed over 10 months. After data collection, statistical analyses of lay error were conducted including fitting non-Gaussian distribution functions and applying Fitts’ Law to the targeting analysis. Experimental results show that (a) the lay error changes with target type and is greater on closer targets and lower on farther targets for the majority of the subjects who participated in the study; (b) The firing time for most shots is below 10 seconds for most subjects, independent of the test conditions, distance to target, or target type; and (c) although the subjects’ shots were mostly accurate, subjects occasionally had significant errors (outliers) that merit deeper consideration.

Soft robots are increasingly gaining popularity as a preferred choice for designing medical devices due to their inherently compliant and adaptable nature. As research continues to advance in this field, soft robots are poised to revolutionize healthcare by providing safer, more patient-centric solutions. This paper presents the design of a novel soft corrugated actuator to create a system for positioning the head during head and neck cancer radiation therapy. Simulation and experimental data show that the presented actuator design can achieve linear expansion with respect to pressure in both free and under loading conditions. A machine learning model of the single actuator is created using experimental data (pressure, expansion, and weight) to predict the required pressure. The single actuator model’s validity is confirmed by using four actuators to achieve head flexion at a specific angle, illustrating its potential for implementation in a multi-actuator system. The preliminary design of the multi-actuator head and neck positioning system demonstrates its capability to achieve four degrees of freedom.

This paper presents an initial investigation of a speech-based human-robot interaction system for locating items in a warehouse environment. The system uses a 2D SLAM map and visual servoing with deep learning-based barcode recognition to identify and locate items based on user speech commands. The system was tested with and without item location in the SLAM map and achieved a 100% success rate in identifying and localizing items. The average speech processing time was recorded at 9.28 seconds, and the system demonstrated a best-case timing of 15.46 seconds and a worst-case timing of 3 minutes for identifying items on different tables. The proposed system has the potential to improve the employment opportunities and experiences of blind or visually impaired workers in the warehouse industry. Future work will focus on testing the system in real-world environments and improving its performance in cluttered and dynamic settings.

In this paper, we develop a novel and safe control design approach that takes demonstrations provided by a human teacher to enable a robot to accomplish complex manipulation scenarios in dynamic environments. First, an overall task is divided into multiple simpler subtasks that are more appropriate for learning and control objectives. Then, by collecting human demonstrations, the subtasks that require robot movement are modeled by probabilistic movement primitives (ProMPs). We also study two strategies for modifying the ProMPs to avoid collisions with environmental obstacles. Finally, we introduce a rule-base control technique by utilizing a finite-state machine along with a unique means of control design for ProMPs. For the ProMP controller, we propose control barrier and Lyapunov functions to guide the system along a trajectory within the distribution defined by a ProMP while guaranteeing that the system state never leaves more than a desired distance from the distribution mean. This allows for better performance on nonlinear systems and offers solid stability and known bounds on the system state. A series of simulations and experimental studies demonstrate the efficacy of our approach and show that it can run in real time. Note to Practitioners —This paper is motivated by the need to create a teach-by-demonstration framework that captures the strengths of movement primitives and verifiable, safe control. We provide a framework that learns safe control laws from a probability distribution of robot trajectories through the use of advanced nonlinear control that incorporates safety constraints. Typically, such distributions are stochastic, making it difficult to offer any guarantees on safe operation. Our approach ensures that the distribution of allowed robot trajectories is within an envelope of safety and allows for robust operation of a robot. Furthermore, using our framework various probability distributions can be combined to represent complex scenarios in the environment. It will benefit practitioners by making it substantially easier to test and deploy accurate, efficient, and safe robots in complex real-world scenarios. The approach is currently limited to scenarios involving static obstacles, with dynamic obstacle avoidance an avenue of future effort.

This article presents a time-invariant extremum seeking controller (ESC) for nonlinear autonomous systems with limit cycles. For this time-invariant ESC, we propose a method to prove the closed-loop system has an asymptotically stable limit cycle. The method is based on a perturbation theorem for maps, and, unlike existing techniques that use averaging and singular perturbation tools, it is not limited to weakly nonlinear systems. We use a typical example system to show that our method does indeed establish asymptotic stability of the limit cycle with minimal amplitude. Utilizing the example, we provide a general guide for analytic computations that are required to apply our method. The corresponding Mathematica code is available as supplementary material.

The ability to quickly detect and identify objects floating on the surface of the water can benefit numerous fields including environmental monitoring, civil infrastructure, and security applications. To address this need, we investigated the ability of multiple imaging sensors to perform machine learning-based object detection and classification for small objects on water. The sensors tested in this effort include a long-wave infrared polarimeter camera, a visible wavelength optical camera, and a low-light camera. Small waste objects such as trash bags, Styrofoam cups, wood, plastic bottles, cardboard, and aluminum cans were placed into a riverine environment on the water surface and data were gathered with each of the sensors. Artificial neural networks were trained based on the data gathered, and models were created to perform object detection and classification of those same objects in the riverine environment. The three-camera systems performed well but with clear advantages for specific environmental conditions and object types.