Bacterial metabolic engineering shows great promise for sustainable chemical production. Non-model microbes such as Pseudomonas putida, Rhodobacter sphaeroides, and Rhodopseudomonas palustris offer unique opportunities for metabolic engineering, given their tolerance to environmental stressors, their ability to grow on waste substrates, and their natural production of industrially relevant compounds. However, tools for engineering these bacteria are underdeveloped. Here we present genome engineering and gene regulation tools that are generalizable to multiple non-model microbes, offering improved versatility for metabolic engineering. Firstly, we employed a high-efficiency genome engineering tool using serine recombinases (SAGE) in R. sphaeroides and R. palustris. We evaluated integration efficiency for 10 different recombinases using a fluorescent reporter screen, revealing variation in recombinase performance across microbial hosts. We used BxbI, the top-performing recombinase, to integrate a heterologous metabolic pathway into the genome of R. palustris for the bioproduction of a biofuel precursor. In addition to genome engineering tools, we developed gene regulation tools using dCas13, a protein which regulates genes at the translational level. Genome-wide functional screens were conducted in P. putida using an inducible guide RNA system to study levels of gene regulation in native aromatic biosynthesis pathways. Overall, this work advances tools for genomic integrations and gene regulation in non-model microbes, offering new strategies for metabolic engineering and expanding the host range for synthetic biology applications. 

Robotic manipulation systems operating in diverse, dynamic environments must exhibit three critical abilities: generalization to unseen scenarios, multitask interaction, and spatial memory. While significant progress has been made in robotic manipulation, existing approaches often fall short in addressing memory-dependent tasks and generalization to complex environmental variations. To bridge this gap, we introduce SAM2Act, a multi-view robotic transformer that leverages multi-resolution upsampling and visual representations from large-scale foundation models. SAM2Act achieves a state-of-the-art average success rate of 86.8% across 18 tasks in the RLBench benchmark, and demonstrates robust generalization on The Colosseum benchmark, with only a 4.3% performance drop under diverse environmental perturbations. Building on this foundation, we propose SAM2Act+, a memory-augmented architecture inspired by SAM2, which incorporates a memory bank and attention mechanism for spatial memory. To address the need for evaluating memory-dependent tasks, we introduce MemoryBench, a novel benchmark designed to assess spatial memory and action recall in robotic manipulation. SAM2Act+ achieves competitive performance on MemoryBench, significantly outperforming existing approaches and pushing the boundaries of memory-enabled robotic systems. Project page: sam2act.github.io.

Additive manufacturing, particularly 3D printing, often produces surface ridges, especially for complex geometries, that require post-processing to achieve a smooth finish. Laser ablation is an effective technique for smoothing these surfaces, but precise identification of ridges is crucial for optimizing the process. This study explores the use of machine learning to detect and ablate 3D print ridges, improving the accuracy of laser smoothing. A convolutional neural network (CNN) was trained on greyscale images of printed surfaces, learning to segment ridge regions from background material. From there, image processing filters and a line transform was applied to gather line defining information to be converted into DXF, a readable file for the laser software. The trained model was integrated into a graphical user interface (GUI) to automate ridge detection and guide the laser for targeted ablation, minimizing manual intervention. The system was validated on test parts, demonstrating overall efficiency and accuracy in ridge identification. Other experiments were done to determine proper laser and process parameters to achieve an accurate and smooth surface finish. The experimental results showed improved surface uniformity. The automated approach made laser smoothing efficient and scalable for industrial and manufacturing applications. By leveraging machine learning, this method advances the precision and repeatability of post-processing in 3D printing, reducing labor costs and improving final product quality.

6PPD-Quinone (6PPD-Q) is a toxic transformation product of the tire rubber additive, 6PPD, that has been identified as the primary cause of Coho Salmon (Oncorhynchus kisutch) mortality in watersheds impacted by roadway runoff. Recent studies have focused on quantifying the lethal concentration of 6PPD-Q, identifying the major sources, and predicting the environmental release from rubber products. Organic chemical release from solids is typically evaluated with solvent extraction where organic solvent and solid are contacted, releasing the leachable chemicals for measurement. However, different solvents and methods introduce inconsistencies in leaching data from different laboratories. This study evaluates the impact of solvent choice on 6PPD-Q extraction from crumb rubber. I will quantify 6PPD-Q concentrations in methanol, ethyl acetate, or acetone during storage after rubber extractions. Determining the best solvent for 6PPD-Q that promotes the most recovery and stability is essential for data quality. After this study, desorption and resorption rates of 6PPD-Q onto various crumb rubbers will be measured.  These studies aim to improve study design for leaching assessments and enhance our understanding of the persistence and mobility of 6PPD-Q in the environment.

In the field of tactile robotics research, two main kinds of sensors are used to capture information, the GelSight and DIGIT. However, the data collection process is extremely expensive and time-consuming. As such, many projects choose one type of sensor to explore claims and develop understanding. Though this may work in the short term, transferring results between the two sensor modalities is still difficult and unclear. My project aims to bridge this gap. Based on the motivation that both sensors capture the same information, I hypothesize that it is possible to seamlessly transition between using them. In order to test this, I developed an embedding model that is able to represent both GelSight and DIGIT data in the same way. Specifically, when both sensors capture the same information, the model is expected to map them to extremely similar outputs. Furthermore, by grounding this embedding model with current unsupervised learning techniques in computer vision, meaningful information is captured in the model's output for use in other tasks. Various model architectures and grounding techniques are tested. They are also evaluated on various downstream tasks to measure the model's usability. Through this work, I aim to allow projects using either sensor modality to share results and collaborate efficiently, closing a gap that has long existed in the field.

Restoration practices are crucial to the sustainable management of city parks, constructed wetlands, and natural ecosystems that have been disturbed or invaded. Oftentimes, restoration sites have some level of disturbance, such as soil contaminants in urban parks. Therefore, selecting plants for restoration comes with a list of considerations based on the goal and scale of the restoration project. Commonly, plants transplanted into these disturbed or polluted environments experience shock from transplanting stress, making finding solutions that increase restoration planting success invaluable to these practices. Soil fungi and bacteria have potential to boost the success of these efforts through their synergistic interactions with each other and plants. These microorganisms have high potential for use as biofertilizers in place of conventional nitrogen- and phosphorus-based fertilizers, which both have negative environmental impacts, including greenhouse gas emissions and water contamination. We hypothesize that by enriching plants by encasing these beneficial bacteria and fungi in alginate-based hydrogel beads, both plant biomass and overall fitness would improve. Further, this improved fitness has the potential to increase post-transplantation survival rates for plants used in restoration and/or phytoremediation regimes. To determine the effect of hydrogel biofertilizers on early stage development and transplant success in a contaminated restoration site, we are examining the response of blanketflower (Gaillardia aristata) to our novel biofertilizer. This plant is rapid-growing, used in restoration, and is drought-tolerant. Therefore, we are pursuing two questions: 1) How does our mixed-consortium hydrogel impact early development of these plants in greenhouse conditions; and 2) Does transplant survivability increase when planted in contaminated soils? Based on previous studies showing the strong efficacy of hydrogel-encapsulated biofertilizers, we predict that plants treated with biofertilizers will have better outcomes (improved early-stage growth and higher survival rates post-transplant) due to their supplemented nutrient accessibility and accelerated growth and development in early adolescence.