Mark Payne

🧠 Memory isn't an add-on—it's the core of any truly intelligent agent. In my upcoming blog, I dive into the actual implementation of a layered memory architecture for AI agents—designed to handle fast recall, session continuity, and long-term knowledge grounding using a triad of: 🔹 L1 – In-Memory Cache (Active Context) 🔹 L2 – Vector DB (Session Memory) 🔹 L3 – Graph DB (Knowledge Memory) This isn't just theoretical. I’ve put this into practice with real agent frameworks and explored how memory impacts performance, continuity, and contextual reasoning. From agent personalization to retrieval-aware decision making, memory is what makes agents feel less like tools—and more like intelligent collaborators. 🚀 The blog walks through each layer, how to wire them up, and the tangible benefits they unlock when combined. Stay Tuned #AI #GenAI #RAG #AgenticRAG #AIAgents

43

2 Comments

Based on the response to the dance, I wanted to share a little behind-the-scenes moment from just a couple hours before showtime. This was the first full run with both robots performing together. Until then, I had only tested sections of the routine with Rivet. Astro had never attempted any choreography before. Same moves, same commands… but very different results. Rivet handled it without issue. Astro collapsed. Turns out he was still compensating for a payload that had been removed earlier. Once we updated the config, he was dancing like a pro. If there’s one thing programming a robot to dance teaches you, it’s how complex its normal motion really is. From CG shifts to footstep sequencing, body rotation, and timing—choreography mode strips away the some of the safety layers and gives you a whole new respect for how these robots move… and even more for what living things can do. #Robotics #EngineeringFails #BostonDynamics #MotionControl #RobotChoreography

109

3 Comments

My latest Substack is now live: Slacker Index: Why vertical integration is key for rapid hardware development In part 2 of a series, I break down what Slacker Index is and why optimizing lead times is so critical for enabling rapid, agile hardware development. https://lnkd.in/gwZczwdf

89

11 Comments

OpenVSP 3.46.0 Released A few features and a bunch of bug fixes. Good stuff all around, everyone should update unless you're stuck on 3.44.X for some reason. If you are, you should try this out to see if it addresses any of the problems you've had migrating. 3.44.X is not going to receive further updates. https://lnkd.in/gvFsjr7r

71

4 Comments

"Be not the first by whom the new are tried, Nor yet the last to lay the old aside." -Alexander Pope I haven't commented much about AI here, but the quote above pretty much sums up my current feelings on the subject. I've watched curmudgeons say all AI is terrible or dangerous and just about shouldn't be used at all. On the other hand, I've seen AI fan boys act like anything not AI is a waste of time. Truth is somewhere in the middle, methinks. AI is not (or shouldn't be) an end to itself. It's a tool. A potentially powerful tool, yes. But just a tool -- and one among many. As another saying goes, "He who is good with a hammer tends to think everything is a nail". No tool is optimal for every need. The part we don't seem to have settled on is where exactly AI fits in our respective toolbags. To that end, I'm glad we have the early adopters trying to stretch it and break it to see what it's really good for. Once a consistent path emerges, I'm happy to travel it too. Use whatever tool seems most appropriate to you for your task. But, if your work affects something like flight safety, you own your work output regardless of tool used. In a post-incident review, it won't be acceptable to say "But, but, but... ChatGPT (or Copilot, Claude, Galaxy, etc.) said...."

11

Below is Part 7 of my white paper series on EVTOL technology development titled "DESIGN TO FAILURE IS NOT AN OPTION, BUT A REQUIREMENT". For all aerospace products low weight is critical, but nowhere is weight more critical than on VTOL aircraft. In an ideal world, all VTOL aircraft parts would have infinite life. In the real world however, limited part lives and margins of safety must be balanced against weight and cost, otherwise VTOL aircraft would be too heavy and too costly to be practical. Therefore, with VTOL aircraft, designing parts to failure at a predictable life is not an option, but a requirement. In this paper I provide a general overview “Life Limited” parts on commercial VTOL aircraft and how they impact operating cost and dispatch reliability. I leave it to the reader to draw any comparisons to potential EVTOL aircraft. In place of links to earlier papers, Parts 1 to 6 are included for easy reference at the back of Part 7. Also, thanks to all you have read my papers and recommended changes   As always, the opinions in my papers are solely my own, and I am open to making any corrections to historical facts (and grammatical ) errors.  Enjoy!

35

2 Comments

Hiring great engineers and researchers? We look for two things: they’ve solved something truly hard - and can apply that thinking to our space fast. Here’s how we test for both: 💻 Deep technical ownership We ask: “What’s the hardest technical problem you’ve solved?” Then go deep for 10–20 minutes. A strong sign: they can explain it simply - but also go so deep into the problem and solution that they lose us halfway. Only someone who’s actually done the work can go there. ⚡ Fast problem-solving We present a real challenge we’re facing. Great candidates explore ideas with us, ruthlessly prioritise, and move fast. That ability to reason in new domains is what startups demand - because the next challenge is always around the corner. These signals are almost impossible to fake.

57

4 Comments

How do you get a 19-story rocket stage from a launch pad to a landing barge hundreds of miles downrange? Navigation systems, of course! In simple terms, a navigation system consists of a sensor suite and software working in parallel to estimate the “state” of a system. For aerospace vehicles, the state we care about is represented by a 1-dimensional matrix called the state vector (x using state-space representation), consisting of vehicle position, velocity, attitude, and body rates in an inertial reference frame. The inertial piece is important because we need a non-translating, non-rotating reference to compare our vehicle dynamics back too. For a rocket, this is a combination of an Earth-Centered Inertial (ECI) and body-fixed frames. The physical device we use for measuring vehicle dynamics is called an Inertial Navigation System (INS). It’s a self-contained piece of hardware containing accelerometers and gyroscopes that measure linear and angular accelerations. Why would we care about measuring accelerations, you say? Because we’re measuring a dynamic system which follows the laws of classical mechanics (i.e., Newtonian and Lagrangian)! If I can measure how my system is changing with time, then I can use my knowledge of system dynamics + differential equations to solve for my state vector! And if I want to refine my solution even further (instead of relying solely on dead-reckoning with routine ephemeris updates), I can integrate my inertial measurements with GNSS measurements, incorporating additional measurements for range, range rate, and PRN (basically a timing variable). Combining all these independent measurements together (with associated error terms), creates our sensor fusion suite! The last piece of this puzzle requires a mathematical method to fit the system dynamics of our rocket stage to the suite of measurements we’re collecting. This is where our friend thr Kalman Filter comes into play, which may also be referred to as the navigation filter. A Kalman Filter is a least-squares regression technique for fitting sensor measurements to a system dynamic’s model (that just means it minimizes the error between the Current-Best Estimate (CBE) of the model and new observable). The true wonder of this method is how it integrates statistics, system dynamics, and noisy sensor measurements to predict an updated estimate of the vehicle state with coresponding covariance matrix. All you need now is navigation software and a processor that can handle the matrix algebra! So in a nutshell, that is how you navigate a rocket stage for a precise landing on a recovery barge! This cycle of collect sensor measurements, pass through the navigation filter, and update the state vector happens many times a second (10 to 100 Hz depending on navigation requirements). And here’s the incredible part of this technique: with this setup, we can estimate vehicle position with sub-centimeter accuracy and velocity with sub-mm/s accuracy!

75

5 Comments