150813 research outputs found
Sort by
1500W High Voltage DC-DC Converter for Electroaerodynamic Aircraft Applications
No full textElectroaerodynamic (EAD) propulsion is a novel form of propulsion that is nearly silent and has no moving parts. The first functional untethered heavier-than-air EAD aircraft had an endurance of 90 seconds and could only fly in a straight line. To enable a practical fixed wing EAD aircraft that can fly outdoors with a payload for an extended period of time, improved power conversion technology is necessary. Prior work specifies a practical EAD aicraft as one with an endurance of 10 minutes, a payload capacity of 200 g, and full controllability. This work explores methods of increasing the specific power of power converters for EAD aircraft from 1.15 kilowatts per kilogram to over 2.0 kilowatts per kilogram. Such an increase can be achieved by utilizing magnetics integration and thermal management techniques, as well as adjustments in the operating point of the power converter. The power converter for the first generation EAD aircraft had an input voltage of 200 V, an output voltage of 40 kV, an output power of 600 W, a specific power of 1.15 kilowatts per kilogram, and an efficiency of 85 percent. In this work, a power converter with an input voltage of 200 V, an output voltage of 20 kV, an output power of 1476 W, a specific power of 2.7 kilowatts per kilogram, and an efficiency of 96 percent was demonstrated to work for a 40 second duration. At the end of the test, device temperatures continued to increase, so it has not been proven that the converter can work in thermal steady state as required for a 10 minute flight. Future work would involve modifying the test setup to allow for adequate ventilation of the ambient air around the converter, as well as modifying the converter with adequate thermal management so as to enable operation under thermal steady state.S.M
Leveraging large language model embeddings to enhance diversity and mitigate the filter bubble effect in recommender systems
No full textThe “filter bubble” describes the potential for internet personalization via algorithmic curation to effectively isolate individuals from a diversity of perspectives or content. In particular, this filter bubble effect may appear as a result of recommender systems used on social media platforms or online marketplaces to influence user behavior. While customization may improve user retention and decision quality, the filter bubble may hinder discovery and intensify polarization, while also reducing the degree of interaction between individuals and different viewpoints or domains. This thesis explores mitigation strategies for the filter bubble by enhancing recommendation models using content-based embeddings produced by large language models (LLMs), which encode semantic information about items being recommended. The addition of semantic information — beyond the user interaction data that usually drives recommender models — may not only improve the quality of recommendations but also promote diversity by allowing for content-based comparison of item candidates for recommendation. After establishing a baseline collaborative filtering recommendation model and validating standard re-ranking diversification techniques, we introduce two LLM embedding-enhanced approaches. The first is a hybrid retrieval scheme that combines collaborative filtering scores with LLM embedding similarity to generate candidate items. The second employs the LLM embeddings directly in a diversity-oriented re-ranking framework. To ensure generality, the same experiments are repeated and evaluated across three widely-used recommendation datasets from different domains. We further explore how embedding granularity influences performance by generating several sets of embeddings encoding different levels of detail and repeating these experiments. We also assess whether a contrastively fine-tuned LLM designed to emphasize inter-item differences produces more suitable embeddings for encouraging recommendation diversity. We reveal that score-fusion hybrids yield negligible diversity gains, particularly on sparse datasets, whereas applying re-ranking shows promise in bursting the filter bubble. In particular, LLM embeddings combined with re-ranking achieve the highest semantic diversity and long-tail novelty across domains and items, at relatively minor losses in precision and other relevance measures.MN
Formulation and Calibration of CATKE, a One‐Equation Parameterization for Microscale Ocean Mixing
No full textWe describe CATKE, a parameterization for fluxes associated with small‐scale or “microscale”ocean turbulent mixing on scales between 1 and 100 m. CATKE uses a downgradient formulation that dependson a prognostic turbulent kinetic energy (TKE) variable and a diagnostic mixing length scale that includes adynamic convective adjustment (CA) component. With its dynamic convective mixing length, CATKE predictsnot just the depth spanned by convective plumes but also the characteristic convective mixing timescale, animportant aspect of turbulent convection not captured by simpler static CA schemes. As a result, CATKE candescribe the competition between convection and other processes such as shear‐driven mixing and baroclinicrestratification. To calibrate CATKE, we use Ensemble Kalman Inversion to minimize the error between 21large eddy simulations (LESs) and predictions of the LES data by CATKE‐parameterized single columnsimulations at three different vertical resolutions. We find that CATKE makes accurate predictions of bothidealized and realistic LES compared to microscale turbulence parameterizations commonly used in climatemodels
Development of a Hierarchical Reflexive Control Framework for Autonomous Robotic Manipulation
No full textWithin the field of robotic manipulation, much research focus has been placed on improving perception and planning algorithms, assuming that the actions output by these high-level planners will be easily achieved by the robot systems. However, to surpass human manipulation performance, fast and robust execution of manipulation plans is just as critical as improved perception and planning methods. In this thesis, we introduce the last centimeter problem, which states that the most difficult part of grasp execution is when less than a centimeter remains between fingertips and an object, and contact is imminent. To solve this problem, we propose a reflexive control framework, which is a manipulation control architecture that decouples low-level, high-bandwidth behaviors, which we call reflexes, from broad high-level plans. The reflexes are fast, autonomous reactions to local sensing information that are designed to add robustness to high-level manipulation plans while also reducing the necessary complexity of manipulation planning problems. To deploy our reflexes, we design hardware platforms that incorporate high-bandwidth actuation and low-latency tactile sensing, allowing us to maximize the reactive capabilities of the overall manipulation system. We validate our approach through studies on teleoperated grasping and autonomous planar grasping, which show that our reflexive controllers increase manipulation speed and robustness. Then, we perform extensive simulation studies for autonomous grasping in SE(3), conducting experiments with single objects as well as cluttered scenes, using a variety of state-of-the-art grasp planners. Our results show greatly improved grasp robustness with our reflexive controllers, across all object types and grasp planners. Further experiments show that the benefits of our reflexes persist across sets of objects that are larger, heavier, and more slippery, and with increasing magnitudes of errors in the executed grasp poses. While this thesis demonstrates that the reflexive control framework is effective at increasing grasp robustness during picking, our framework is constructed in a way that is amenable to extension to other tasks, like in-hand manipulation or constrained object placement, as well as application to more complex grippers, such as those with three or more dexterous fingers and more diverse sensing.Ph.D
Report to the President for year ended June 30, 2025, Laboratory for Nuclear Science
No full textThis report contains the following sections: Experimental Nuclear Physics, Experimental Particle Physics, Institute for Artificial Intelligence and Fundamental Interactions, Theoretical Particle and Nuclear Physics, MIT-Bates Research and Engineering Center, MIT Central Machine Shop, and Education
Graphics4Science: Computer Graphics for Scientific Impacts
No full textSIGGRAPH Courses ’25, Vancouver, BC, CanadaComputer graphics, often associated with films, games, and visual effects, has long been a powerful tool for addressing scientific challenges—from its origins in 3D visualization for medical imaging to its role in modern computational modeling and simulation. This course explores the deep and evolving relationship between computer graphics and science, highlighting past achievements, ongoing contributions, and open questions that remain. We show how core methods, such as geometric reasoning and physical modeling, provide inductive biases that help address challenges in both fields, especially in data-scarce settings. To that end, we aim to reframe graphics as a modeling language for science by bridging vocabulary gaps between the two communities. Designed for both newcomers and experts, Graphics4Science invites the graphics community to engage with science, tackle high-impact problems where graphics expertise can make a difference, and contribute to the future of scientific discovery. Additional details are available on the course website: https://graphics4science.github.io
Fisher–Schultz Lecture: Generic Machine Learning Inference on Heterogeneous Treatment Effects in Randomized Experiments, With an Application to Immunization in India
No full textWe propose strategies to estimate and make inference on key features of heteroge-neous effects in randomized experiments. These key features include best linear predic-tors of the effects using machine learning proxies, average effects sorted by impact groups,and average characteristics of most and least impacted units. The approach is valid inhigh-dimensional settings, where the effects are proxied (but not necessarily consis-tently estimated) by predictive and causal machine learning methods. We post-processthese proxies into estimates of the key features. Our approach is generic; it can beused in conjunction with penalized methods, neural networks, random forests, boostedtrees, and ensemble methods, both predictive and causal. Estimation and inference arebased on repeated data splitting to avoid overfitting and achieve validity. We use quan-tile aggregation of the results across many potential splits, in particular taking mediansof p-values and medians and other quantiles of confidence intervals. We show thatquantile aggregation lowers estimation risks over a single split procedure, and establishits principal inferential properties. Finally, our analysis reveals ways to build provablybetter machine learning proxies through causal learning: we can use the objective func-tions that we develop to construct the best linear predictors of the effects, to obtainbetter machine learning proxies in the initial step. We illustrate the use of both infer-ential tools and causal learners with a randomized field experiment that evaluates acombination of nudges to stimulate demand for immunization in India
Robust Dexterous Manipulation Enabled by Learning at Scale inSimulation
No full textRobots with robust bimanual dexterity have the potential to transform industries such as manufacturing and healthcare by performing complex tasks at human-level proficiency. While end-to-end learning methods have shown promise in achieving this goal, scaling these approaches remains challenging. Existing paradigms suffer from high costs associated with collecting large-scale, high-quality demonstrations on physical systems and face performance saturation due to reliance on offline data. We propose a task-agnostic pipeline that leverages robotics simulation to overcome these limitations. In particular, we introduce DART, a cost-effective, augmented reality, robot teleoperation platform for scalable data collection. We demonstrate through user study that it enables twice the throughput of existing systems. We also present a learning algorithm that integrates real-world demonstrations with reinforcement learning to surpass performance plateaus. Finally, we design a method that zero-shot transfers policies trained in simulation on real robots using only RGB input. Together, these contributions provide a practical and scalable path toward achieving general-purpose dexterous robot manipulation.M.Eng
Report to the President for year ended June 30, 2025, Department of Materials Science and Engineering
No full textThis report contains the following sections: Undergraduate education, Graduate education, Graduate and postdoc career support, Student organizations, Facilities, Fundraising, Personnel changes and promotions, Research highlights, Awards and honors, and Future plans
An Interactive Visual Paradigm for Knowledge Graph Question-Answering
No full textIn an era of information overload, verifying data reliability and provenance is critical, yet knowledge graphs (KGs) often remain complex for non-expert users. This thesis introduces TRACE, a Reasoning and Answer-path Comprehension Engine, a visualization tool enhancing transparency in KG question answering (KGQA). By abstracting intricate KGs into intuitive meta-nodes, TRACE simplifies exploration of large, multi-topic datasets. Its interactive interface allows users to navigate semantic communities and trace reasoning paths, fostering trust through clear answer derivation. Unlike cluttered traditional graph visualizations, TRACE’s meta-node approach provides a scalable, user-friendly solution, concealing technical complexities while enabling robust query validation. Large language models support natural language query parsing and community summarization, making KGs accessible to diverse audiences. TRACE positions itself as a vital widget for information platforms, empowering users to counter misinformation confidently. A user study and pipeline evaluation confirmed TRACE’s intuitive interface excels for complex queries, though multi-hop paths pose challenges, while processing tests demonstrated its scalable paradigm for large datasets. By prioritizing transparency and usability, TRACE redefines KGs as reliable tools for knowledge discovery, laying a foundation for future systems to deliver trustworthy, accessible information in a digital landscape fraught with uncertainty.M.Eng