41530 research outputs found
Sort by
Framework for Multi-Agent Coordination and Distributed Localization in Micro-UAVs
Get PDFMicro-UAVs (unmanned aerial vehicles) due to their inexpensive nature and compact form factor have shown an increase in prevalence throughout a multitude of applications including, but not limited to: search and rescue, military reconnaissance, and agriculture monitoring. However, for a majority of these high impact applications, a swarm of micro-UAVs are required and furthermore mandate that they are able to cooperatively and autonomously coordinate with each other. For long, controlling and communicating between a user and a singular micro-UAV has been a well known and solved problem, however the same can\u27t be said for swarms of micro-UAVs. This project seeks to address this problem by developing a framework for Crazyflies, a type of micro-UAV, to cooperatively communicate and coordinate movements with each other. This is ultimately achieved by utilizing a REST API to serve commands from the user and collect localization data in a pseudo-P2P (peer to peer) fashion between the drones in the swarm. Additionally, the swarm of Crazyflies coordinate movements based off localization data served by the pseudo-P2P network via a leader-follower hierarchal topology. To conclusively validate this approach experiments with three Crazyflies flying in a predefined triangle and line formations are conducted to confirm the reliability and accuracy of the pseudo-P2P network and coordination based off the leader-follower topology. While the drones were generally able to maintain their intended relative positions, quantitative analysis using the RMSE metric revealed measurable deviations, particularly in the line formation, where the average RMSE reached approximately 0.466 meters. Nonetheless, the system exhibits robust real-time performance and reliable formation maintenance, demonstrating its potential for a wide range of multi-agent robotics applications. Overall, this implementation of a framework that enables reliable and fast micro-UAV swarm coordination, opens the door for a multitude of applications such as search and rescue and agriculture monitoring
Investigating and Evaluating Poison Unlearning under Imperfect Detection
Get PDFMachine unlearning is an emerging field focused on removing the influence of specific data points from trained machine learning models. Early work in this field has primarily focused on privacy-oriented unlearning, often motivated by data protection regulations such as the GDPR. More recently, security-oriented unlearning has emerged as a response to data poisoning attacks, where malicious inputs are injected into training data to manipulate model behavior. In these security-oriented settings, unlearning methods aim to reverse the influence of malicious training data after a model has already been deployed. However, existing methods assume access to an accurately identified portion of poisoned data. This thesis investigates the performance of state-of-the-art poison unlearning techniques under imperfect detection conditions, where the forget set is not only a portion of poisoned data but also includes false positives. An evaluation simulating these more realistic detection scenarios is designed. State-of-the-art unlearning method, Potion, is evaluated using this framework across standard image classification benchmarks. The results demonstrate that false positives in the forget set can impact model damage and unlearning effectiveness. More broadly, these findings highlight the need for robust and comprehensive evaluation frameworks that reflect real-world conditions
Simulated Live Studio Audience
Get PDFThe Simulated Live Studio Audience is a Python based application that utilizes Vosk, Roboflow, and Llama 3.2 to provide a user with auditory feedback based upon both visual and audible input from their device\u27s microphone and camera. This system functions with a custom trained computer vision model to detect a specified object and when individuals walk in and out of the camera frame, outputting sitcom style simulated crowd reaction sounds accordingly. The simulated studio audience program also takes in vocal input from users, converts it to text, and, using a large language model, analyzes it for content that can be interpreted as sad or humorous, and produces an auditory crowd reaction appropriately
Spy Bot
No full textThis project set out to develop a compact, mobile-controlled robotic platform designed for remote video surveillance and control via a smartphone. Using a Raspberry Pi as the core controller, the robot integrates a Flask-based server, HTML/JavaScript-based user interface, and an infrared-capable camera to provide real-time video streaming and directional control through a web browser. Key design challenges included managing power delivery, integrating motor control with live video, and modernizing web technologies to ensure smooth communication between client and server. The final system demonstrates a responsive and portable proof of concept that highlights the potential for future upgrades, such as night vision capabilities, autonomous behavior, or enhanced maneuverability. The project offers a solid foundation for exploring mobile robotics with an emphasis on accessibility and real-time interaction
Electroporation Protocol Development for Cal Poly Cell Therapy Laboratory Courses
Get PDFCancer is a heterogeneous disease caused by unchecked cell proliferation. The disease contributed to 9.7 million deaths worldwide in 2022, with hematological malignancies (leukemias, lymphomas, and myelomas) accounting for over 1.4 million deaths in the United States from 1999 to 2020. This thesis focuses on the treatment of these cancers, with an emphasis on non-Hodgkin lymphomas (NHLs). R-CHOP (Rituximab-Cyclophosphamide, Doxorubicin hydrochloride, Oncovin/Vincristine, and Prednisone) is the standard of care first-line treatment for diffuse large B-cell lymphoma (DLBCL), the most common subtype of NHL, and encompasses a mixture of chemotherapy, immunotherapy, and corticosteroids. The R-CHOP regimen cures 50% to 70% patients, leaving 30% to 50% of patients with either relapsed or refractory (R/R) disease. For patients who are unresponsive to R-CHOP, high-dose (salvage) chemotherapy and autologous stem cell transplantation (HDC-ASCT) are typically the second-line treatment option. Patients who are resistant to these first- and second-line treatments are candidates for cellular immunotherapy, such as chimeric antigen receptor T cells (CAR T cells), as a third line (and sometimes second-line) treatment. A CAR consists of a single-chain variable fragment of the human IgG antibody, traditionally targeting CD19 or B-cell maturation antigen (BCMA), a costimulatory domain, and CD3 chains, allowing a patient’s autologous T cells to recognize and kill cancer cells that have evaded the immune system. Despite the promise of these therapies, the lack of skilled personnel required to scale therapy development and the need for more precise CAR transgene insertion stand as obstacles in advancing this field. The Cell Therapy curriculum at Cal Poly aims to address these deficits by teaching students critical techniques relevant to cellular therapy development. The work in this thesis aims to improve upon a protocol for T cell electroporation to introduce students to a simple transfection model (electroporating GFP into Jurkat T cells) relevant to cellular therapy development. Electroporation can deliver the Cas9 plus gRNA system into cells for precise gene insertion, allowing this protocol to also lay the groundwork for future CRISPR transfection for stable gene expression. Because preliminary electroporation experiments yielded low cell viability (13% in a 10 μL electroporation reaction volume), I aimed to increase post-electroporation live cell percentages. I compared varying concentrations of the GFP plasmid payload, electroporation pulse parameters, and electroporation reaction volumes to determine the combination that yielded the highest cell viability with sufficient GFP expression. These experiments suggested that a GFP plasmid payload of 75 μg∙mL-1, a 100 μL electroporation reaction volume, and the initial recommended electroporation parameters from Thermo Fisher (1 pulse, 1700 V, and 20 ms pulse width) are the optimal conditions to increase post-electroporation live cell percentages over 3-fold from the preliminary experiment. Future work should investigate the longevity of GFP expression and explore transfecting a transgene and plasmids containing Cas9 plus gRNA to induce stable gene expression
Comparative Study of Experimental and FEA Strain Data for CNT-Coated Fiberglass Sensors, Extensometers, and Metal Foil Strain Gauges
Get PDFMaterial testing is essential across industries such as aerospace, automotive, and construction, playing a critical role in verifying material selection, diagnosing failures, and understanding the development of flaws in structures. These insights are key to designing successful, reliable systems. Conventional metal foil strain gauges are low cost and reliable but provide limited sensitivity with a typical gauge factor around 2. Extensometers provide highly sensitive strain measurements with the disadvantage of a bulky form factor. With advanced materials such as carbon nanotubes, it is possible to manufacture a sensor with the sensitivity closer to that of an extensometer with the small form factor of a metal foil strain gauge. This thesis developed a strain sensor utilizing the piezoresistive properties of carbon nanotubes, and compared it with conventional metal foil strain gauge, extensometer, and finite element analysis data for uniaxial tensile and cantilever beam tests. The CNT sensor was found to be 82% more sensitive to strain resulting in a 5.65% higher Young’s modulus than extensometer and FEA data during tensile testing of a fiberglass specimen. Additionally, an optimized CNT sensor was found to be 160% more sensitive than a metal foil strain gauge and 48% more sensitive than this study’s initial CNT sensor for aluminum cantilever beam tests. Double cantilever beam (DCB) tests were performed with metal foil strain gauges resulting in strain within 8.09% of FEA and hand calculations, with many of the samples within 2.09%. The optimized CNT sensor is expected to be within 4% of this data when applied to the DCBs, similar to results of this study’s aluminum cantilever beam test
The Impact of Accessibility Features on Player Experience in Video Games
Get PDFWhile video game accessibility is a growing research topic, few studies investigate how players perceive the presence versus the absence of accessibility features, or how non-disabled players react to the option of accessibility features. This study explores these research gaps, investigating how access to accessibility features affects the experience of both disabled and non-disabled players. For the purposes of this study, a small platformer game was developed with as many accessibility features as feasible for the scope of the project. An A vs.\ B study was conducted in the game, with anonymous participants randomly assigned to version A, with all acceptability features enabled, or version B, which had most features disabled. Overall reception to the features was primarily positive in both groups. A small portion of non-disabled players made use of the accessibility features. Most disabled players assigned to Version A used the accessibility features. Non-disabled players were primarily indifferent to the accessibility features, but some found them helpful or liked them even if they did not need them. Though strong conclusions are difficult to draw without more data, overall accessibility features are significantly desired by disabled players, and found to be useful and received positively by some non-disabled players. Further research is needed, but results show potential that including accessibility features can positively influence audience reception
Seconds From Impact: Anticipatory Vehicular Crash Prediction Using Video Vision Transormers
Get PDFVehicular collisions represent a significant public health concern, necessitating re search into advanced emergency notification systems. While deep learning has shown promise in accident detection, a research gap persists in applying state-of-the-art transformer architectures to the task of anticipatory, real-time crash prediction from video. This thesis addresses this gap by developing and evaluating a Video Vision Transformer (ViViT) for the binary classification of imminent vehicular collisions. Utilizing a curated dataset of 1,493 unique collision sequences, this study systemati cally investigates the impact of temporal context by comparing the ViViT against a single-frame Vision Transformer (ViT) baseline and conducting comprehensive exper iments on temporal hyperparameters like tubelet depth and frame stride. The results compellingly demonstrate that leveraging temporal context yields substantial per formance improvements, with the optimal ViViT model achieving 98.72% accuracy and, most critically, a recall of 98.75%—a 7.63 percentage point improvement over the baseline. The findings validate the efficacy of pure attention-based models for this safety-critical application and establish a strong methodological foundation for developing intelligent transportation systems capable of reducing emergency response times and saving lives
Effect of Household Oil Contamination and Agricultural Soil Contamination on Phyllosilicate Nanocatalyst Performance During Pyrolysis of Polyolefins
Get PDFThe effect of household oil and agricultural soil contamination on catalyst performance during polyolefin pyrolysis was simulated on mixed virgin plastic pellets by evaluating contamination influence on mixed plastic degradation temperatures. Three phyllosilicate nanocatalysts including Fulcat435, Fulcat22F and Y-Zeolite were evaluated in this work based on a literature review of catalyzed plastic pyrolysis and these catalysts’ ability to lower plastic decomposition temperature by 50 °C – 100 °C. These findings were confirmed in this research through thermogravimetric analysis of a polyolefin mixture and further developed in furnace experiments containing household oil and agricultural contamination. Household oil contamination deactivated each of the three catalysts evaluated through blockage and poisoning of catalysts active sites. An increase in household oil contamination was directly correlated to a decrease in catalyst performance. As contamination ratio increased, the degradation profile further deviated from the uncontaminated baseline and statistical significance (α = 0.05) increased. It is anticipated, after a contamination ratio of 0.20 (plastic : contamination), the effect of household oil contamination plateaus and degradation profiles would match those similar to uncatalyzed pyrolysis. Experiments performed with Y-Zeolite and Fulcat435 were approximately two times more susceptible to household oil induced catalyst deactivation compared to experiments performed with Y-Zeolite and Fulcat22F. These results indicate that catalysts with smaller surface areas and or greater acid values are less susceptible to catalysts deactivation due to household oil contamination. It is proposed, the tighter phyllosilicate structure of Fulcat22F prevented interaction with larger fatty acid molecules of household oil contamination. Additionally, the greater acid value of Fulcat22F is directly correlated to an increase in active sites composed of Brønsted and Lewis acids. The author proposes these active sites provide extra capacity to catalyze reactions, further suspending complete catalyst deactivation. Although complete deactivation is possible, greater contamination ratios would be required to saturate active sites within Fulcat22F. To confirm this hypothesis, future work should identify potential correlation between phyllosilicate acidity and susceptibility of catalyst deactivation to household oils. The effect of agricultural soil contamination on catalyst performance is uncertain based on the results presented herein as oxygen present in agricultural soil induced combustion reactions increasing plastic degradation. An evaluation of the gas fraction of degradation products is recommended for future work to quantify combustion reactions initiated by oxygen present in agricultural soil contamination. Agricultural soil contamination present in plastic pyrolysis may provide a benefit as carbocation radicals generated from combustion reactions and could further catalyze plastic degradation. However, combustion reactions should be mitigated to preserve the quality of gaseous products. A pre-washing process should be performed if catalyzed pyrolysis is performed to mitigate contamination entering the pyrolysis furnace and maximize catalyst efficiency due to vulnerability of catalysts deactivation from contaminants commonly present in the municipal plastic waste stream
Underground LRT Station
Get PDFThis report presents a comprehensive fire safety and evacuation systems analysis for the underground light rail terminal station project. The station features two main levels— concourse and platform—and is strategically located adjacent to a sports stadium.
Station Overview: The underground light rail terminal station is a two-level facility strategically positioned adjacent to a major sports stadium. The station operates as a critical transportation hub with distinct functional zones designed for efficient passenger flow and emergency evacuation:
Concourse Level (3.5 meters below ground): Functions as the primary passenger processing and circulation hub. This level serves as the main entry control point where passengers purchase tickets, pass through fare gates, and transition between street level access and train platforms. The concourse accommodates passenger waiting areas, retail spaces, and technical support rooms on both sides. Three escalators and four emergency stairwells provide vertical circulation to the platform below, while direct street-level access through northern and southern entrances enables rapid evacuation to the point of safety.
Platform Level (10.1 meters below ground): Houses the train boarding operations with an island platform configuration serving bidirectional rail traffic. This level manages passenger queuing, train boarding/alighting operations, and platform safety through platform edge doors (PED) that control access to trains. The platform includes passenger waiting areas and emergency circulation via four emergency staircases (1.85m width each) that provide direct egress routes to the concourse level above.
Technical Support Infrastructure: Distributed across both levels, technical areas house critical building systems including electrical equipment, HVAC systems, fire suppression infrastructure, and the Station Control Room (SCR). The SCR provides backup control capabilities to the main Operations Control Centre during any emergency event.
Emergency Systems Integration: The station features comprehensive safety infrastructure including automatic sprinkler coverage, smoke detection throughout all areas, emergency communication systems, and dual 250-cubic-meter water reservoirs that supply fire suppression for both station and tunnel operations. Four emergency stairwells (1.85m width) provide protected egress routes with 120-minute fire-resistant separation.
Operational Capacity: The station is designed for normal operations with up to 1,339 passengers, with specialized crowd management protocols for high-occupancy events reaching 2,200 passengers during adjacent stadium events.
The project addresses several unique challenges: (1) Responsibility for providing fire suppression water supply to the tunnel line through two 250m³ water reservoirs, (2) Integration of a secondary control room serving as backup to the Operations Control Centre (OCC), and (3) Management of extreme fluctuations in occupancy during stadium events.
The design incorporates comprehensive safety systems: (1) EST3 standard automatic fire detection and suppression system with complete coverage and 276-second response time, (2) Smoke extraction system with 14.16m³/second capacity in each of two ventilation shafts, (3) Four emergency staircases with a 1.85m width and three escalators for evacuation, and (4) Reinforced concrete structure with 6cm concrete cover for enhanced fire resistance.
Comprehensive engineering analysis confirmed compliance with NFPA 130 standards, demonstrating a platform evacuation time of 2.57 minutes and total evacuation time of 4.47 minutes. Computer simulations verified that at the designed occupancy of 1,339 passengers, all safety systems function properly and enable safe evacuation.
Critical Findings - Performance Comparison: Prescriptive vs Performance-Based Analysis Results
Platform Evacuation Time: 2.57 minutes (Prescriptive Analysis) Total Evacuation Time for 1,339 passengers: 4.47 minutes (Prescriptive Analysis) and 4.0-5.9 minutes (Performance-Based Analysis Evacuation Time for 2,200 passengers: 7.57 minutes (Performance-Based Analysis) NFPA 130 Compliance (6-minute limit): 4.47 minutes (Prescriptive Analysis). Exceeds at maximum capacity (Performance-Based Analysis)
Summary and Recommendations: The comprehensive engineering analysis emphasizes the importance of maintaining the 1,339- passenger occupancy limit as a central parameter in station safety. The combination of engineering solutions (detection systems, fire suppression, and smoke extraction) with operational solutions (Crowd Manager, emergency protocols) provides an optimal response to the project challenges.
The underground light rail terminal station project demonstrates an effective integration of advanced engineering design and efficient operational management, enabling it to address the unique challenges of a terminal station while ensuring compliance with all standards and regulations, maintaining passenger safety in all possible scenarios