August w3 2024: Busy week this week - we demonstrated scalability of packing sims up to the size of a knee-high sock (Our largest test yet!), it should be printing over the weekend. In order to automatically position linkmeshes in the build volume, we built a proxy simulation that finds optimal starting orientations without user intervention. In the course of doing more research for our Q-Submission to the FDA, we learned that we are closer than we thought to being able to start the study. Additionally, we found a software provider that may allow us to take scans of the patients using nothing more than an iPhone, which clinicians can do easily at scale, rather than using a bulky piece of specialized hardware. We are also looking other cloud providers that can meet our growing GPU needs - if anyone has good availability for the L40S, please drop us a note! August w2 2024: We worked on developing the simulation architecture for larger (>10k links) linkmeshes - lo and behold, we also need large GPUs. Currently, faster GPUs give us faster sims, but we are also working on ways to balance compute load across multiple GPUs, while updating rough collisions in a shared memory, so that we can pack multiple linkmeshes into a single print volume. We also received the first draft of our FDA Q-Sub, which we will be refining over the next week or so. August w1 2024: We printed two new test articles (a glove for one of our adivsors, and a sphere), which validated that the printer we plan to use for production during the study does in fact maintain nominal quality below certified specs. In order to get there, we ran many sims and fine-tuned the parameters for this particular printer. We also archetected and began development for a larger orchestration system that meets and exceeds the maximum level of data privacy required by HIPPA/FDA for patients/users, while negligibly changing development and user experience. We also made a clean scan of June's leg, which we'll use for testing leg-scale garments. July w4 2024: We continued to tune our new simulation system; we're now getting better results from it faster. We just ordered our first print from the new pipeline, and we're going to be fabricating a lot of test hardware next week. We met with one of the only spacesuit patternmakers in the world, who also has MCP experience, and had a super interesting meeting. With the pressure sensor matrix, we have all the equipment needed to quantitatively test our designs, so we hope to share some results soon. July w2 2024: Solved nearly all problems around printability, we should be printing hardware version 2 gloves next week. We made some big updates to our cloud backend, which will give us more control and flexibility than our previous cloud operations system - we are now able to use GPU accelerated simulations in production. Got potentially good news on the FDA front - clearance might be faster than we were expecting, but too early to say for sure. We are expecting the pressure sensor array mid-week tomorrow, so hopefully we can get a print through to test it out with as soon as it arrives! July w2 2024: Our simulations now run ~10x faster and are more stable, but we uncovered a few challenges around the printibility of our models. Most of the fixes were implemented, but some of the more fickle problems are still under development. We purchased the pressure sensor array, which we will use to be able to start getting quantified pressure data out of our testing. This timing great, as pressure data is critical to understand during our next main engineering focus of closure design and stiffness programming. July w1 2024: Closed out the refactor of the back side of the manufacturing pipeline. New advantages include: standardized methods for control over link size, shape, simulation parameters, meshing parameters, and tracking location over time. This foundation can also now be trivially expanded for future mesh developments. We also did quite a bit of research into additional manufacturing methods that may allow us to manufacture in metal, but they are very niche. Would definitely prefer to figure out how to optimize the strength of SLS and MJF parts. June w4 2024: Worked on knowledge transfer over to our FDA consultant, as well as implementing our new rigid body techniques. Our benchmarking is showing that models that previously took single-digit days to simulate, could now be done in hours, opening the door to much faster manufacturing turnarounds than we previously thought possible. We are also looking at moving offices for more customizable space for manufacturing. If anyone is interested in sharing some office space in the BNY, email me. We also learned that NASA's spacesuits, and spacesuit development deals, are falling apart - it is our sincere hope that we can be part of the solution. June w3 2024: We unblocked quite a few things: rigid body physics on the GPU (developed an (as far as we are aware) novel system for doing broadphase collision detection on the GPU, which according to Nvidia Devs, may make its way into their native collision engine), outsourcing of FDA device research and writing, and we found a great solution for rapid turnaround on SLS manufacturing by using a local underutilized printer. June w2 2024: Worked on a few different fronts, but all towards creating software version 2, enabling hardware version 2. Translating between the computational geometry code of software 1 to the rest of the stack on version 2 has been a headache, but we have a great strategy now. A big part of hardware version 2 was sourcing the matrix pressure sensor, but we are still up against manufacturing turnarounds, which I think will be a focus for the next week or two. We also got more solid timelines on FDA clearance for the study - slower than we were hoping for, but that time will be put to good use, and it doesn't change our mid-term timelines. June w1 2024: Implemented the multicore rigid body physics sim, though we expect a few fine-tuning steps to take going forward. Went to an interesting startup science fair as part of NYC's tech week, where I met a lot of interesting people. I also got to try out a different type of flexible matrix pressure sensor array, though I'm not sure this one is right our our needs due to a number of factors. Additionally, our sample finished printing, but I won't be able to pick it up until Monday. May w5 2024: We made great strides on a transition over to a multicore rigid body simulation framework, which will allow for denser packing, faster. We also made good progress on a refactored computational geometry framework, which will increase our interation speed going forward, but still more to do on that front. We also chatted with a fantastic regulatory consultant who is aiding on FDA filings for our clinical study, made improvements to our office space, and ordered a fabric sample that was generated using the new code! May w4 2024: Lot of progress on multiple fronts this week: We investigated high-resolution pressure sensor arrays for evaluating compression methods, found a new rigid body architecture that has historically handled incredibly good chainmail simulations, took our first steps on a refactor that will allow for more control over our prototyping and manufacturing in the future (and interfaces well with the new physics engine), handled additional admin setup to allow Opensuit to handle R&D expenses and bookkeep more easily, and got the ball rolling on a non-provisional. May w3 2024: Automated the now-containerized computational geometry -> physics simulation -> manufacturability post-processing pipeline, iteration speed is now wildly improved and variables are isolated. We also started moving towards a new physics engine, which will require work up front, but will yield more accurate simulations, faster. I also took care of quite a few admin tasks following our raise. May w2 2024: Closed our Angel II round! Big thanks to Boost VC and 1517, as well as all individual angels, all helping to make the next generation of spacesuits a reality! We also welcomed June Knauth to the team - she will be working on software engineering, computational geometry, and whatever else needs to happen! Since she started, she's restructured and containerized my computational geometry algorithms, which can now run in the cloud with complete horizontal scalability. May w1 2024: Did a lot of logistics and admin work around closing this round, ultimately settled on raising a larger amount that will allow us to hire 1-2 more people, carry out at least one clinical study for the compression garments, and make spacesuit prototyping happen concurrently with medical development. I also had two great meetings with the astronaut, and with a veteran of the medical device market. April w4 2024: Did a massive amount of outreach this week - I connected with a NASA astronaut, one of the only living capital-E Experts on mechanical counter-pressure, a new mentor, and new investors. While I've been doing this administrative work, June has been working on restructuring (not completely refactoring just yet, though that will come) the codebase so that we can iterate much faster using AWS' HPC rather than my laptop. April w3 2024: Got two prints back this week - one is a fantastic test article, and highlights a few areas that we need to work on for physics sims. The other is pretty much hard as a rock. Crazy enough, both are MJF printed using the same model (different materials, but that shouldn't change it this much). Need to work on manufacturer relationships, and some torus packing math. On the business side, I started Opensuit's fundraise! I've been doing a lot of outreach and polishing up a deck, so if you want to chat, shoot me a note by adding o@ to the site. April w2 2024: Got a signed LOI from the hospital! They are going to help run the study, and become customers if all goes to plan. We also built extension to the QBAC office so that we have room for anticipated team growth. In the second half of the week, I went on a team retreat with 1517, where I met some of the smartest people I now know, got incredible insight and advice on my work, and spent some time not in front of a computer. I expect to get my printed prototype back when I arrive back in NYC. April w1 2024: Through talking to one of my manufacturers, I found a geometry bug in the translation step between simulated geometry and printed geometry, and fixed that. I am getting ducks in a row for the study - evaluating manufacturing relationships, ensuring sure everyone at the hospital is on the same page about the composition of the study, finding additional advisors to help who's experience will help with different aspects, and evaluating the cost of the research. June and I worked on containerizing the computational geometry and simulation code, so that it can all run on the cloud (and not be reliant on my personal computer!), which will also be instrumental in increasing manufacturing cadence when we start the study. I'm also considering getting a small SLS setup for opensuit so that we aren't reliant on external providers. March w4 2024: Visited the clinic, and got great feedback and reception on the glove prototype I brought in. I've built relationships with a few key people at the hospital, and we have a meeting set to plan out a study. Also, fully implemented overpacking checks on both the statistical evaluations and the final generation, so I can now be almost entirely confident about manufactureability. March w3 2024: Provisional patent on inelastic compression garment fabric has been filed. On the manufacturing side, at least one of my prototypes may be nonfunctional, due to the specific machine having worse tolerances than anything I planned around. It's my fault for not confirming that, hopefully the other MJF and SLM prototypes will have better tolerances. If not, find a new manufacturer (or process). I also did a lot of amount of research onto FDA class 1 registration, and feel confident that I can do that when the time comes. Next week, I'll be going to the hospital again to start getting a study scoped out. March w2 2024: Ordered three new models to test materials and the new manufacturing code - one MJF Polypropylene, one MJF Glass-Filled Nylon, and one SLM Aluminum. Solved a few problems in sim stability, which is now at an acceptable state for large scale integration. I also started a high level overview of the process to aid in refactoring (potentially by a good friend, potentially by me). Patenting is moving ahead but taking longer than expected, will be just in time for conversations about FDA clearance with SMEs. March w1 2024: Wrapped up the computational geometry! My parallel processing algorithm is nearly completely functional and certainly helpful, but not perfect. As of right now, everything is in place to generate printable meshes with no/minimal manual fine-tuning. I think the optimal ratios between FOL min/maj and SOL packing haven't been settled on yet, so I'm going to continue revising that. I hope to manufacture another glove by mid-next week. I also started on my provisional patent, that should be settled by mid-next week as well. February w4 2024: Visited a wound care clinic, and got a completely overwhelming amount of interest. Was entirely unprepared for people to just start asking when they could try it! Also had some fantastic meetings about what the next stage and seed raising will look like. One the code side, I worked on getting more predictable generation across multiple meshes to try to avoid having "magic values" that overfit for the meshes I've been testing with. Did this by making a simplified version of SOL generation that allows me to predict SOL minor radii for a given coverage ratio - this will greatly improve manufacturability, and already works pretty well, but has a little way to go before I can trust it completely. February w3 2024: Busy week- I'm moving offices as the work transitions from digital to physical, and the need for space grows. A lot of this week was spent on setup logistics, leases, and insurance, but did get to have some great meetings additional medical applications for custom fit compression garments, and what expected turnarounds would be. I also started to get a new glove manufactured, but learned that some of my recent optimizations may have pushed me a tad too far outside what can be manufactured, so I worked on a montecarlo method to simulate manufacturing steps in parallel. Still needs work but feels promising. February w2 2024: Finished the complete end to end generation pipeline for this stage. Physics simulations are still manually started and ended, but there's still unknowns in terms of simulation iterations, so I'm comfortable with it. Should at least be good enough for small trial runs of treating lymphoma or venous stasis/ulcer patients, so I'm gearing up for that. If I get more software development bandwidth, I found some new avenues for physics sims which could be worth exploring, but that would probably mean hiring someone. I'm also about move office spaces for more space to grow and test, but this week had a fair bit of paperwork because of that. February w1 2024: Bulk of this stage of engineering work is over, I'm now starting work to move into implementation out in the world. I'm beginning with clinical trials for inelastic compression garments at Brooklyn Hospital, and spacesuit-based fashion/luxury goods. A few of the highlights from wrapping up engineering work included a 30% speed improvement across four different optimizations, on paper worked out how programmable extensibility could work, found that my numpy idx -> blend idx code was off in some edge cases and fixed it, and automated most of the physics sim -> printable step. Next week I'll need to wrap up the last parts of the physics -> printable (mostly how to handle the closures links), have some meetings about testing in hospitals, and getting on runways. January w4 2024: Solved the closures location and joining problem. It wasn't quite as bad as I thought it would be, but working through the logic of how the closures should handle edge cases, and actually implementing that between py and blender was certainly challenging. Ultimately, getting both sides of my pipeline to agree on which SOL where which was the most tricky, but I got there. I'm now working towards testing as medical compression garments- it's my hope that the programmability and custom fit can make compression garments that are far more helpful for patients than conventional elastic garments. January w3 2024: Finished out programmable rigidity, and made some easy-to-communicate visualizations for it. Given that everything is paint-on, I'm starting to wonder how much of this I could conceivably automate if I had vascular scans (or some other input). I also started on the print-in-place closures at the end of the week. Based on the rate of progress I had starting out there, I expect it to be non-trivial, but we'll have to see. January w2 2024: Did a lot this week - nearly finished out the programmable rigidity, and got to a local maxima with link generation that I'm very happy with. I also fought my battles against git, and won. Earlier in the week, (and more or less for the last few weeks) I've been struggling with some frustration and self-doubt in my abilities, but I feel that I've really broken through that, and am incredibly excited about what's ahead. January w1 2024: Fixed a bunch of overly fragile parts of generation, which are now pretty much across the board better. Lots of things that I thought I had adequate solutions for at the end of December ended up not working nearly as well as I needed them to, but now it's good to go. SOl scaling and shaping is much less unwieldy, and gives me better per-link-system control, rather than only focusing on individual links. Making coarse tori for physim may be more challenging than previous iterations because of this, but the quality of the linking will improve simulation stability immensely in the long run. December w3 2023: Reframing SOL generation ended up being a headache, but one that I expect I will be thankful I took care of now, rather than later. SOL generation now takes the minor diameter of connected FOL into account when selecting the correct FOL from the lookup. December w2 2023: Quite happy with where I got the closure locating logic to, and I did some early hardware prototyping on the sinching mechanism. In the process of reworking the closure locating technique, I wrote an extension of the remeshing algorithm I use that creates meshes that are much more uniform in triangle aspect ratio and edge length (given an input edge length), which I've now adopted everywhere. Additionally, in implementing printable geometry for the closures, I exposed an overly-fragile part of my SOL generation, which frankly I was very surprised to never have encountered before. I got a start on reframing SOL generation, and will be closing that out early next week. December w1 2023: Working extensively on the logic and design for paint-on closures and paint-on rigidity, primarily focusing on closures. Made substantial progress, but nothing printable yet. Parametrization of this problem is non-trivial, and I wholly expect to revisit this down the road when I'm doing rapid wearable testing in conjunction with geometry processing. I also spent some time at the end of the week backfilling all code updates from July w1 2023 to September w2 2022. I have yet to x-ref this with print order history to capture the prototyping history, but that will be easy to add next week. November w4 2023: Early in the week, I worked out a hacky way to paint-on rigidity and closures, and did a bunch of research into using PhysX for parallelized physics. On the rigidity side, the approach I had used would make later iteration tricky, so I'll be revisiting this next week. PhysX is crazy powerful, but I'm going to need to dedicate quite a bit of time to migrating my workflow there, so I'll do that down the road. I spent the rest of the week with my partner's family for thanksgiving, which was wonderful. November w3 2023: Largely worked on the SOL orientation and scaling problem, so this was a linear algebra heavy week! Some lossy design and architecture were excised, leaving an algorithm that creates second order links that are scaled and rotated to perfection, even in very high curvature regions such as between fingers. To demonstrate how well this new approach can handle complex and high curvature geometry, I'm making a utah teapot. November w2 2023: Investigated different surface reconstruction algorithms (VCG from the Visual Computing Lab is fantastic) and overhauled how I calculate curvature weights for Lloyd's algorithm. I was frustrated by how little control I felt that the current implementation gave me, so I'll likely make my own in the future. I also uncovered a pretty essential flaw in how my SOL placement worked, and started to address it. November w1 2023: Got an incredible scan of my right hand done by NYCAP3D - full color with 5M polygons! I could probably pull my fingerprints off the model. This time, the scan is of a hand in a half-closed position, as most (all?) spacesuit gloves are manufactured. I primarily did infra work, cleaning up and processing the new test model, and getting remote file storage for the hundreds of 3d models generated in testing, as per Johnny's suggestion. October w3 2023: Attended ASCEND 2023 in Las Vegas, met really great people and had interesting conversations! Implemented genuine version control for my work - looking forward to the ease of record keeping, shareability, and parity across devices. Figured out how to get insurance, and made a substantial breakthrough on increasing the predictability of computational geometry algorithms by continuing to refine the way I approach the unitary units information loss problem. October w2 2023: Implemented the segmentation edges as weights for the surface packing algorithm, but it didn't work as well as I was hoping it would. I suspect that some of the techniques could be useful when working on LoNEs in a few months. I also spent some time struggling with scale inconsistencies across filetypes, which was subtly messing with some of my algorithms but is now solved, and got some cards and such to prep for ASCEND in a few weeks. October w1 2023: Developed a 3D-mesh segmentation algorithm that identifies different segments of a mesh (eg. individual fingers, palm from wrist from back of hand), and made it run far faster than I thought was possible. I think this will allow me to bias the mesh surface packing algorithm with better precession around fine structures, such as between the fingers. September w4 2023: Identified one of the primary areas of complexity in development as the non-uniformity of units across steps of my algorithm. This partially can't be helped because many 3d model file formats are Unitary Unit, and there aren't great existing libraries for handling 3MF files. September w3 2023: Completed first wearable and fully articulating glove. Will work on characterizing the metamaterial, improving fit, and reducing areas of structural loss due to simulation instability. September w2 2023: Found that through process of working on scaling and tuning other factors, the weighted voronoi for link size has become fragile. Added programmability to that factor, but this effected scaling in other parts of the pipeline, which now need to be brought back up to code. Placed order for new glove using current pipeline. September w1 2023: Developed code to make variable size links that don't compromise simulation stability or printed flexibility, but it required some extreme workarounds for the physim program to behave properly. Working wih quaternions is tough, but ended up being the hail mary that made everything work due to the lack of gimbal-lock. August w4 2023: After much testing, discovered that the root cause of all simulation stability issues was multi-body intersections in geometry. Due to the way that object generation is implemented in the physics simulation, this is non-trivial. August w3 2023: Investigated alternative directions for rigid body processing - found working solutions, but I expect to restructure physim work from scratch down the line. Discovered an oversight with the incircle/mesh surface packing and projecting algorithm from last week, which got fixed and slightly improved simulation stability and mesh regularity, but simulation instability is still an issue. Also attended Professor Bonnie Dunbar's lecture, which was fantastic. August w2 2023: 3D Voronoi incircle-based spacing math so that first-order tori to be centered as well as possible, and second-order tori can be smaller (thus decreasing metamaterial mesh thickness and increasing flexibility). August w1 2023: Made a new 3D visualization workflow automation to minimize time spend on file-handling and possibility for false-positive errors due to humans doing a machine's work. July w4 2023: Worked through a lot of geodesic distance problems and ultimately wrote my own greedy spacial/graph traversal algorithm because the pure graph-traversal algs (eg. bi-directional Dijkstra, A*) were too finicky to get working with 3D data in the way that I wanted. SOL are now better placed between FOL July w3 2023: Did primarily admin-related work. July w2 2023: Opensuit raised our angel round from 1517 Fund! July w1 2023: Worked on model/metadata/config data handoffs between the python geometry processing environment and the blender geometry processing environment. At the time, this was non-trivial because I had completely outgrown my file-handling organization strategy without realizing it. June w4 2023: Prototyped about a half dozen geodesic walk weighting schemes to try to force better placement of the SOL between FOL. This was always a trade-off between voro processing time (coarser mesh is faster, and anecdotally has better convergence) and SOL placement (finer mesh has more direct geodesics because there are more options for half-edges to traverse) June w3 2023: Experimented within the remesh-density to lloyd relax iterations "phase space" (used very loosely here) to try to find better magic values. June w2 2023: Worked on making the algorithm that defined and placed the plane more robust. As it turns out, simply running RANSAC on poorly structured data gives poorly structured results. Moved over to linear algebra rather than relying on statistical methods. June w1 2023: Developed a plane projection algorithm to try to place the SOL better. Roughly, I was using a geodesic walk to get a point on the surface, but it would usually be skewed a few MM to the left or right of the Euclidian midpoint (if viewed antiparallel to the local surface norms, the euclidian mid wouldn't be attached to the surface) so then I would project it onto a plane between the voro cell centers. May w3 2023 -> May w4 2023: Trying to get better midpoints by making more aspects of the geodesic path and SOL scaling process parametric to account for wide ranges in major diameters of FOL to be bridged. May w2 2023: Continued working on refactoring all the code for the new simulation tolerancing paradigm, and ensuring that the interior "psudo-surface" of the maile is what actually aligns with the scanned model so that fitment is preserved. Some smart friends had raised good questions about this, but I was too in the weeds to think about how to solve this best. I'm pretty satisifed with the solution I implemented, but I suspect later versions will take into account the local biomechanics of the particular area of the body. May w1 2023: Had some great introductions this week, but projects at BMCC took up a lot of my time April w4 2023: Had a realization about tolerancing in simulation while I was in the shower, which dramatically shifted the goalpost in terms of simulation stability, but could hold the promise of prints with no required post-processing! Started work on all the adjustments I'd need to make in order for this to work. April w3 2023: Began implementing all the physics settings parametrically, and ways to simulated rotation and location data out for printing upscaling and printing. April w2 2023: Focused on parametrization more generally, and in making the SOL placement technique more nuanced. April w1 2023: Leaned more into using weighted graph traversal, with more interconnects between the spacial representation of the mesh and the graph representation, particularly for identifying geodesic neighborhoods on a mesh (i.e. sections that an ant could walk along the surface to in equal time). March w4 2023: Made some tools to identify where kinks with the geodesics / graph traversal were, and tried a few naive approaches to solving them. This stretched my 3D transformations and what basic linear algebra I have, so some time was spent simply learning what math existed to handle R3 projections and planes. March w3 2023: Moved to lean more on graph traversal algorithms as geodesics for SOL placement information, which yielded a rapid improvement in maile quality. March w2 2023: Following some advice I got from an investor friend, I worked on making my deck more data-backed and professional, including some hefty data analysis of every single usage of a spacesuit to date. February w2 2023 -> March w1 2023 Made money to continue supporting myself and this development. February w1 2023: Frustrated with the quality of the Blender simulations and the general iteration speed of this project, I looked for low-hanging-fruit ways to increase iteration cadence, and discovered Pypy. January w4 2023: Continued working on blender, with an emphasis on improving simulation stability. January w3 2023: Blender generation comes with a lot of benefits but the growing pains are really tough, because blender doesn't give me as much flexibility to rapidly "hotswap" scaling or rotation code without restarting the application to clear all vertices out of memory, and is clearly built for just about every other use case. It feels like I'm trying to use a jet engine as a lawn mower motor. November w3 2022 -> January w2 2023: Did branding and outreach work, but mostly needed to focus on paying the bills and being with family. November w2 2022: Investigated ways to increase the testing cadence of the lloyd algorithm, tested other sculptural applications for the mesh, and moved away from pure-python mesh generation and towards blender generation. November w1 2022: Realized that I had been overfitting for the one model of my hand I was using, so I started testing out the mesh generation algorithm on various sculptures and other 3d models. I was, in fact, very overfit, so I took some steps towards generalizing. One of the ultimate goals was to start use cases beyond spacesuits, and at this point, intricate inflatable structures feels like a good avenue. October w4 2022: Worked on some new visualization and tracking tools, worked on further optimizing the voro weighting algorithm, and ramped up work on the communication and public facing side. October w3 2022: Began using surface weights tied to curvature in the lloyd relaxing step to get better placement and scale for the FOL. For whatever reason, this broke the voro cell adjacency technique I was using before, but I wrote a new adjacency-finding that I'm pretty proud of. October w2 2022: Refined the SOL torus generation, employing a rudimentary approximation of a 3D quadratic Bézier curve for more accurate placement between FOL voro centers. This gave better, but far from perfect, alignment and orientation for the SOL. Also, made some run duration and readability improvements . October w1 2022: Characterizing the SOL alignment problem, wrapping my head around the plethora of ways that this can be wrong. Took a stab at offsets for SOL in tight angles. September w4 2022: Started working with graph algorithms, which allowed me to make the first voronoi-derived chainmail meshes. September w3 2022: First passes at generating oriented tori that are aligned and sized to match patches (voro cells) on a surface, really just getting my feet wet with rotations of meshes in R3. September w2 2022: Generating non-oriented torus meshes (all vertex matrices and face matrices), and writing a "flood fill" algorithm for voro cells on the mesh based on vertex adjacency. Code from this period is lost due to SSD failure. The zenbook pro duo does many things well - thermal regulation is not one of them. From this point forward, all code exists with three copies, is regularly backed up, and is never stored in the same physical location. However, regular pruning of old code (which in my case, was quite bad) is to Code following Akin's 11th law