DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Hybrid cognitive robotics+ & explorations therein with the robot peri.2
No full textJuly2025School of ScienceIn \textit{cognitive robotics}, all substantive actions on the part ofa robot, both physical and mental, are a function of reasoning over
the robot's beliefs directed at declarative
propositions, where the propositions are represented as formulae in the formal
structure of some logical system, and the reasoning is precise deduction defined
in the logic's \textit{proof theory}. Joined by colleagues, I have
expanded via three steps the discipline of cognitive robotics as
defined by Levesque by allowing: (i)
cognitive attitudes beyond belief to be included
(e.g.,~\textit{knowing}, \textit{intending}, \textit{desiring}), as
long as these attitudes are directed toward content expressed in
formulae in the relevant logic (as in the case of
\textit{belief}); (ii) reasoning to be non-deductive; and (iii) the
content in formulae to be non-declarative (e.g., importantly,
imperative). My work in this expanded approach to cognitive robotics
has already met with considerable success, as shown by the initial
prowess of Perception Enabled Robotic Intelligence 2, a cognitive
robot known simply as `PERI.2'. However, it is essential to note that robotics and AI have recentlyreceived much attention because of advances not in logic-based
techniques, but because of the success of deep- and
reinforcement-learning techniques and their application to big data.
This provided adequate reason to question a strictly logicist approach
in cognitive robotics. In response, I adopt and find encouraging
results in a hybrid approach to cognitive
robotics in which I marry the logicist approach of (i)--(iii) with
approaches like deep learning; thus, \textit{Hybrid Cognitive Robotics Plus} (HCR).
Deep learning, as is well-known, eschews manual engineering based on
reasoning over structured, declarative data; HCR combines such
engineered and logicist techniques with sub-symbolic
and data-driven techniques (e.g.\ Machine Learning). The fact is,
while sub-symbolic reasoning is powerful in some domains
and for some problems, there are numerous applications where it remains
unacceptable due to a desire for procedures requiring precise reasoning and
programming, and solutions that rely on
such reasoning and programming:\ e.g.,~safety, reliability, and formal
verification. On the other hand, a swathe of problems are simply
beyond the reach of logicist techniques, for instance robust image
recognition. The research approach that drives this dissertation is
the development and use of hybrid techniques that integrate logic-based
algorithms with sub-symbolic processing. Using these techniques to increase the capability of PERI.2 will bethe main thrust of the dissertation. My extension of the foundation
erected in cognitive robotics in working with colleagues will include
building out deeper theory, achieving markedly better implementations
in prior application areas, and engineering implementations in some
new application areas. This better implementation is the application of deep logical reasoning with perception to object manipulation to solve complex
problems; use of a physical robotic manipulator through the
development and deployment of PERI.2 expanding on prior works featuring the incomplete robot. Overall, the robot's use of perception with logic remains unique among its peers. Previously un-attempted application areas are solved through this technique; the key specific challenge is solving physical logic puzzles. A social deduction challenge was also solved. Additional challenges were unmet but substantiate the work, including sculpting based on literary prompts, working in occluded perception environments, and tasks in in-the-field
emergency medicine (e.g.,~applying splints for bone fractures using
malleable material). Each of these problems involves deep logical
reasoning to reach a solution that is reliable, explainable, and safe
in a real-world environment. Each problem benefits greatly from
sub-symbolic processing due to the need to interpret complex visual
scenes, a traditionally challenging task for logic-based approaches.
While each domain into which my specific challenges fall has been
worked on for a long while, my work pioneers the integration of
cognitive robotics, modern machine learning, and classical engineering
robotics and human-robot interaction. Work aimed at human-level robot
capability currently being carried out by industrial ``heavyweights''
like Google, OpenAI, Boston Dynamics, and Intuitiv is nearly
exclusively sub-symbolic in nature and would, I believe, benefit
greatly from my hybrid approach. Lastly, a significant part of the novelty of my research inheres inthe use of state-of-the-art computational logic; I shall specifically
make use of, and contribute to the refinement of, automated reasoners
and planning systems (in the latter case, the RAIR Lab's Spectra
planner).Ph
Ai models for decentralized finance
No full textAugust2025School of ScienceThis dissertation develops a general framework for modeling user behavior from transactional data, using decentralized finance (DeFi) as a motivating case study. DeFi protocols such as Aave offer a rare opportunity to study large-scale, real-world financial behavior through publicly available transaction-level data. However, this data is complex. It is high-dimensional, heterogeneous, and irregularly timestamped. To address the modeling challenges this poses, this dissertation explores a range of methods, including clustering, survival analysis, transformer-based representation learning, and code generation with large language models. We begin by characterizing user behaviors in Aave using quarterly address-level summaries and unsupervised clustering, revealing dominant behavioral archetypes and their evolution over time. Building on this, we introduce a novel survival analysis framework tailored to DeFi, modeling event timing from raw transaction sequences and uncovering patterns in loan repayments, liquidations, and platform usage.
These insights motivate the creation of FinSurvival, a benchmark suite of 16 large-scale survival prediction and classification tasks. FinSurvival is the first publicly available benchmark of its kind in finance, and we show that standard machine learning methods often outperform deep learning models in this high-censoring environment. To explore learned representations, we develop Large Transaction Models (LTMs). These transformer-based models generate embeddings of transaction sequences. We evaluate the effectiveness of these embeddings on FinSurvival tasks, demonstrating that they can improve performance in classification settings relative to both raw and hand-engineered features. Finally, we present a benchmark for evaluating the ability of LLMs to generate code for analyzing transaction data, demonstrating the feasibility of natural-language-driven automation of data querying and transformation. Collectively, this work contributes new methods, datasets, and evaluation tools for behavioral modeling with transaction data. It highlights the challenges of modeling with transaction data, the tradeoffs between hand-engineered and learned representations, and the promise of AI models for handling this kind of data.Ph
Two-stage chromatographic purification of lentiviral vectors with enhanced particle characterization via nanoflow cytometry
No full textAugust2025School of EngineeringLentiviral vectors (LVVs) are central to the advancement of gene therapy, particularly in ex-vivo applications such as chimeric antigen receptor T (CAR-T) cell therapy. However, existing purification methods of LVVs face significant challenges, including process variability, colloidal instability, and low functional recovery. This thesis addresses these limitations through the development of a two-stage, orthogonal chromatographic purification process designed to enhance both recovery and stability of LVVs.Anion exchange chromatography (AEX) remains a foundational unit operation in LVV purification, but the strong electrostatic interactions between LVVs and quaternary amine ligands often necessitates high salt concentrations for elution, which can compromise vector integrity. Additionally, conventional bead-based stationary phases rely on pore diffusion to utilize their total binding capacity- however, the size of LVV inhibits its diffusion into traditional bead-based media. To overcome these limitations, we evaluated monoliths, which enable large modalities like LVV to access the entire functionalized surface area via convective transport, resulting in improved mass transfer, higher binding capacity, and faster processing times, enabling greater recoveries.
We evaluated Arginine hydrochloride (ArgHCl) for its potential to improve recovery and colloidal stability during the CIM QA step. Arginine is a widely used formulation additive known for enhancing protein solubility, suppressing aggregation, and improving elution profiles. ArgHCl substantially improved physical and infectious recovery in linear gradient elution experiments. In addition to improving recovery, dynamic light scattering measurements revealed that ArgHCl enhanced the colloidal stability of eluted fractions compared to the conventional NaCl eluent, where particle aggregation was evident.
Nanoflow cytometry was then employed to characterize particle heterogeneity and assess particle retention behavior of VSVG-positive and VSVG-negative subpopulations across LGE experiments. This analysis revealed a 2-peak elution profile on CIM QA, with a primary peak consisting predominantly of VSVG-negative particles and a secondary peak enriched for VSVG-positive particles, which corresponded to infectious recovery. The tetraspanin protein CD9, a known exosomal marker protein, was found to be enriched in the initial impurity peak of LGE fractions. The impurity fraction containing the highest levels of CD9 was then recombined with infectious particles from the secondary peak, resulting in a measurable reduction in infectivity. These findings suggest that exosomes constitute a substantial portion of the initial impurity peak and may impair LVV transduction efficiency if not effectively removed during purification.
Following AEX development, a secondary flowthrough polishing step using Capto Core 700 was implemented to further reduce host cell protein (HCP) and double-stranded DNA (dsDNA) impurities. Improved LVV recoveries were observed when low concentrations of
ArgHCl (150–300 mM) were added in the feed, without compromising impurity clearance. As a result, the Capto Core step was positioned second to the CIM QA capture stage. When operated in tandem, the integrated two-step process achieved high recoveries of both vector transgene and p24, along with a 2-log reduction in HCP and dsDNA levels. Overall, this study establishes a two-stage purification process that enables effective impurity clearance and high LVV recovery.Ph
Absorptive performance of layered metasurfaces using microslit panels
No full textDecember 2024School of ArchitectureThis work examines the effectiveness of layered metasurface arrangements, prioritizing space efficiency to create a broadband sound absorber. Innovative metasurfaces, including microslit panels with concentrated and coiled cavities, provide frequency-dependent absorptive properties. While these metasurfaces are independently known as efficient sound absorbers, single-layer microslit panel absorbers with a single cavity, in general, lack a wide effective bandwidth. In this work, theoretical models guide the effective creation of metastructures to achieve broadband absorption. These systems are then assembled and measured using an impedance tube to validate their acoustic performance. This paper discusses the formulation of the theoretical model, experimental validation of the model for layered absorbers, and the design specifications that can be met using various metasurface combinations.M
Investigation into aeroacoustic characteristics of evtol rotors
No full textDecember 2024School of EngineeringAdvancements in technologies including batteries and electric drive trains has led to the developmentof a variety of concept electric vertical takeoff and landing (eVTOL) vehicles. In this
thesis, the aeroacoustics of different eVTOL platforms is considered under different conditions,
highlighting the importance of phasing, aerodynamic interactions, and rotor solidity.
A group of classical multicopters is first considered, with a quadcopter, hexacopter, and
octocopter operating at the same conditions compared against an equivalent single rotor vehicle.
These platforms are large enough to be “manned-scale”, meaning that rotors are large
enough to be pitch controlled and thus phase locking between rotors is possible. Initially the
multicopters are considered in hoverwith just two special phasing cases, “orthogonal” and “tipto-
tip”, which are opposite phasing cases in which adjacent rotors are perfectly in or out of
phase. The acoustics are compared to examine how phasing conditions compare on each multicopter
platform, with the orthogonal phasing showing much greater acoustic benefits. The
multicopters are then examined at different cruise speeds using the orthogonal phasing, with
alternative disk loadings also considered. It is found that configurations and disk loadings with
lower advance ratio maintain acoustic signatures fromphasing patterns at higher cruise speeds,
and that locations of advancing blades is a large indicator of high noise accumulation. The
quadcopter is then examined again, this time with additional, higher solidity rotors in addition
to the baseline considered in hover and cruise. Relative phasing is set to 216 possible starting
phases by varying the starting position of three of the rotors against a static rotor. Noise across
all possible phasings is compared, finding that depending on the relative phasing between rotors
noise varies quite significantly. Additionally, the orthogonal phasing considered previously
is shown to be acoustically one of the best phasing configurations for 2-bladed rotors, while the
tip-to-tip is one of the worst acoustically.
Aerodynamic interactions, which can occur due to rotor proximity to other rotors, the
ground, or wings, significantly alters the noise produced. A rotor pair in ground effect is examined,
with two ground spacings that show weak and strong aerodynamic interactions. The
interactions from the ground cause a large increase in the high frequency content of the noise,
and when the interactions are stronger this effect is amplified and the noise signal is significantly
altered. A line rotor pair is then examined in cruise, with the wake from the front rotor
interfering with the rear rotor thrust. Despite the differences in load between a rear rotor in isoxx
lation and experiencing interactions, the acoustics show almost no difference whether or not
the interactions are present. A propeller-wing assembly in axial cruise is examined, with the
interactional effects examined for both the prop and wing independently. The propeller noise
shows large increases in front and behind when a wing is present due to an increase in peak
loading, despite an overall loss in thrust. The wing noise is found to have large high frequency
content, which is more strongly predicted when higher fidelity aerodynamic simulations are
used for noise predictions.
Wind turbines share the same broadband noise models for acoustic predictions that are
used in rotorcraft. Using a methodology that combines rotorcraft and wind turbine acoustic
prediction methods, the acoustics from a single-rotor and quad-rotor wind turbine are compared.
It is found that for observers closer to the rotor plane, the quad-rotor is louder due to
sources being perpetually closer to the observer compared to the single-rotor turbine.
The rotors being considered for eVTOL concepts are typically much higher solidity than
traditional rotors, allowing for a lower tip speed which results in lower noise produced. A baseline
rotor is established, and then the solidity is increased through both an increase in chord
and number of blades. By comparing the acoustics and performance metrics of the rotors, it
is found that increasing solidity through number of blades is best for acoustics, and at higher
solidity values acoustic benefits diminish while performance penalties do not.Ph
Folktale story generation and automatic evaluation of generated text
No full textDecember 2024School of ScienceFolktales, as cultural narratives originating from oral traditions, simultaneously entertain and educateindividuals. These indigenous tales, deeply rooted in societal contexts and imbued with moral
lessons, provide invaluable insights into the customs and traditions of the communities responsible
for their creation and transmission across generations. Consequently, folktales possess historical
and literary importance, making them ideal tools for fostering understanding and promoting cultural
exchange among diverse cultures. Remarkably, folktales from various societies frequently exhibit
striking similarities, highlighting the interconnectedness of human experiences across the globe.
Folklore researchers have meticulously analyzed and classified folktales based on types and
motifs to identify shared themes and characteristics. Nonetheless, the automatic identification
and discovery of folktale types remain an area of ongoing study, and little work has focused on
automatic folktale classification and clustering.
The research presented here has three distinct contributions: addressing the challenges
associated with automatically recognizing folktale plot types and motifs to cluster tales, generating
narratives conforming to one or more established folktale types, and evaluating the produced text
with regard to the plot type and a holistic measure of quality.
This research lays the foundation for valuable applications in diverse fields, including information
retrieval, persuasive communication, negotiation strategies, natural language comprehension,
and computational creativity. Additionally, the capacity to abstract natural language semantics could
be crucial in various cognitive tasks, and this study offers insights into these fundamental processes.
Lastly, this research advances a computational perspective on cultural influences, enabling the
exploration of cultural distinctions as they manifest within narratives.Ph
Pegylation of staphylococcus aureus protein a ligands for improved efficiency in monoclonal antibody chromatography purification
No full textDecember 2024School of EngineeringStaphylococcus aureus protein A (SPA) affinity chromatography is extensively used in the purification of monoclonal antibody (mAb) therapeutics due to the specific interaction between the SPA ligand and the mAb Fc region, providing effective removal of impurities generated during the cell culture production phase of the process. However, the resulting eluate must still undergo further treatment through a series of polishing chromatography steps in order to adequately reduce impurity levels to an acceptable range specified by regulatory agencies before preparation of the drug substance and drug product. Conjugation of SPA ligands with polyethylene glycol (PEG), a non-toxic, inert, highly tunable, and water-soluble polymer, through a process called PEGylation has been previously studied in the context its potential to reduce non-specific binding events and its impact on mAb binding. The PEGylated SPA media was observed to reduce host cell protein content while retaining mAb yield during host cell culture fluid (HCCF) purification, increasing the selectivity of the resin for mAbs. SPA ligand PEGylation strategies were investigated in more detail in this work to gain a better understanding of the relationship between the excluded volume generated by the conjugated PEG and the binding behavior and purification efficiency of the PEGylated SPA media. In the first aim, characterization of commercially available CaptivA PriMAB SPA media was conducted to assess the changes in physical properties and binding behaviors of resins after PEGylation through the N-terminal conjugation of linear PEG chains ranging in size between 21.5 and 41.4 kDa, along with a branched 20.5 kDa PEG chain. PEGylation resulted in the alteration of the pore size distribution of the media as a function of PEG size, and varying impacts on the antibody binding capacities of the resins. HCCF purification with respect to host cell protein and residual DNA removal was most efficient with the 21.5 kDa PEGylated resin, although the amount of eluted aggregate content trended inversely with the volume of conjugated PEG per ligand. The 39% increase in selectivity for mAbs observed from HCP content reduction for the 21.5 kDa PEGylated resin compared to the unmodified media reinforces the viability of PEGylation as a means to improve the selectivity of SPA media for mAbs.
The second aim addresses the use of an alternative PEG attachment site. A novel SPA media based on a Z-domain dimer with an A104C point mutation was conjugated through a thiol-maleimide reaction to assess the impact of the PEG binding site on media antibody binding characteristics and HCCF purification efficiency. Maleimide activated PEGs differing in size between 5.4 and 20.6 kDa were assessed. The 5.4 kDa PEGylated media generally exhibited minimal impact on the binding capacity and dissociation constant of the media. Increasing conjugated PEG size resulted in reduced extents of PEGylation, but reduced the antibody binding capacity of the media and the antibody binding strength to the ligand. The 5.4 kDa PEGylated media was most effective in reducing impurities during HCCF purification while retaining reasonable mAb yields, indicating that the extent of PEGylation reaction may play a significant role in the reduction of nonspecific binding events. When comparing the 5.4 kDa PEGylated media to the unmodified media, a 98% increase in selectivity for mAbs through HCP content reduction and a 53% increase in selectivity for mAbs through rDNA content reduction are both observed with the PEGylated variant, further supporting PEGylation as a strategy to effectively improve selectivities of SPA medias.
In the third aim, molecular dynamics simulations of PEGylated B domains of SPA were used to gain insight into the orientation and space occupied by PEG and the polymer’s impact on the binding region of SPA on a molecular level. Simulations were conducted using linear 5 kDa, linear 20 kDa, and branched 20 kDa PEG chains attached at the N-terminus of SPA, as well as a linear 5 kDa PEG chain attached to the thiol group of a mutated B domain variant of SPA from alanine to cysteine at the 46th residue (A46C) which mimics the mutation investigated in Aim 2. Results indicated that the increased linear PEG size exhibited larger volume occupancy around the antibody Fc binding region of SPA. Additionally, PEG attachment at the thiol site showed no overlap with the SPA-mAb Fc binding region, reflecting the minimal influence on binding capacity observed in Aim 2. Impacts on the change in Gibbs free energy of the antibody Fc-PEG complex were estimated from the N-terminus PEGylated SPA simulations, while the thiol-targeted 5 kDa PEG revealed no effect due to the lack of occupation of the SPA-antibody binding region.
Overall, this work advanced the understanding of the performance of media with PEGylated SPA ligands, and provides insight into the relationship between the extent of reaction, size of conjugated PEG, and conjugated PEG volume on media performance and impurity clearance for the further optimization of PEGylation in protein A affinity chromatography.Ph
A unified architecture and control framework for safe and collaborative human-robot manipulation of deformable objects
No full textMay2025School of EngineeringDeformable object manipulation (DOM) is pervasive in daily life and in industrial settings. Tasks such as cable routing, tent manufacturing, pouring granular material from bags, and laying up composite sheets all involve objects with high internal degrees of freedom, making them notoriously difficult to model and control. This dissertation focuses on enabling multiple mobile robots, in collaboration with human operators, to perform DOM in various scenarios—ranging from 1D ropes and cables to 2D fabric and composite sheets—while ensuring safety, efficiency, and ease of use. A central challenge in DOM is obtaining accurate state estimates from sensors that are frequently subject to occlusions and limited feedback. To address this, we employ position-based dynamics for real-time simulation of object motion, contact, and friction, providing critical data such as predicted object shape, stress, and proximity to obstacles. This simulation underlies a suite of controllers. First, we incorporate control barrier functions to ensure the robots adapt their motion and maintain safe distances from obstacles and prevent overstretching of the deformable objects.
Second, we develop an efficient global planning pipeline to manipulate deformable linear objects in cluttered environments, approximating them as serially connected rigid links.
Third, we introduce a local control framework for 2D composite layup, where robots transport and position large fabric sheets collaboratively with a human operator. The same simulation reports tensions and contact forces to avoid overstress and prevent unsticking of the material from curved surfaces. We validate the developed architecture and algorithms in simulation and with physical robots under different modes of shared autonomy: human teleoperation, human-robot collaborative manipulation, and fully autonomous control with human guidance. Demonstrations include robotic rope and stiff rod navigation through obstacles, multi-robot formation in tent manufacturing, and real-time composite layup assistance. The system features a unifying touchscreen user interface that simplifies multi-robot programming and visualization, promoting seamless scalability across tasks and facilitating broader industrial adoption. Ultimately, this work advances DOM toward a robust, user-friendly framework, paving the way for safe and versatile human–robot collaboration on a wide range of deformable objects.Ph
Commonsense AI in the History of the Web
No full textMachine common sense (MCS)-the challenge of enabling computers to grasp everyday human knowledge-has been a grand challenge in Artificial Intelligence (AI) since the 1950s. While recent advances in large language models have led to impressive progress, there is still no consensus on how much common sense today's AI actually possesses. In this brief review, we revisit the historical development of MCS in the context of the Web, examining how the Web's evolution-from early knowledge representation efforts to knowledge graphs, the Semantic Web, and crowdsourcing-has shaped MCS research. We argue that key breakthroughs in Web technologies were instrumental in addressing longstanding challenges of scale and coverage in commonsense reasoning. At the same time, MCS research has influenced the development of core Web applications, including intelligent agents, plausibility-based reasoning, and robust evaluation of black-box AI systems
Idle yet engaged: how idle games satisfy our needs for competence and autonomy
No full textJuly2025School of Humanities, Arts, and Social SciencesTraditional theories of player engagement emphasize the central role that challenge and overcoming failure play in sustaining interest, with research showing that gradually increasing challenges play a key role in driving engagement. However, idle games contradict this assumption by generating prolonged periods of play in the absence of mastery, challenge, or failure, thereby exposing gaps in our existing understanding of what drives player engagement. This dissertation aims to examine how idle games, despite lacking these elements traditionally considered essential for long-term engagement, satisfy psychological needs. It argues that Cognitive Evaluation Theory (CET) has been too narrow in its need-satisfaction criteria and has not considered the possibility that these needs can be satisfied not through negative feedback but rather through its absence. It then proceeds to test these claims with a longitudinal study in which participants played one of four versions of an idle game designed to manipulate competence and autonomy. This research provides the first empirical evaluation of the ability of idle games to satisfy a player's needs for competence and autonomy. In doing so, it reveals novel approaches for fostering long-term engagement without negative feedback or increasing difficulty. These findings refine our understanding of player engagement and offer broader implications for game design and gamification.Ph