have a draft blog about robotics benchmarking. Originally was going to talk more about sim eval design but maybe I should pivot to just talk about “what should be benchmarked” some benchmarks are good, but some are just pure distraction atm (eg this particular test)

Ack, will follow up

The best programming language of all time is C Just C

End-to-end whole-body mobile manipulation on Digit — straight from research onto the show floor with zero-shot performance in a chaotic environment. #IROS2026 @agilityrobotics

I wanted to share something I've been working towards since forever. I've finally built out a full slam + hand reconstruction pipeline that works IN REAL TIME at 30fps on a single rockchip 3588. All streaming directly to @rerundotio from a @BitRobotNetwork robocap. This is similar to what a Quest3 or Project Aria 2 provides, but opensource 🙂 I'm so excited I finally don't have to really on closed source software like the meta quest or others. I finally have a realtime performant pipeline that works on an edge computer, but importantly can also scale to a high performance one when needed.

Excited to announce Volantis's $88M Series A. We are solving Al's memory bottleneck by using optics, enabling chips with huge amounts of fast & cheap memory. By boosting both the memory bandwidth and capacity per chip by orders of magnitude, we enable ultra-fast inference (up to 10,000 tps/user) for large models (>10T) - with low $/tok to boot. Initially, this will enable insanely fast agents - think coding agents that finish in minutes or even seconds instead of hours. More excitingly, optics is a fundamentally scalable way to increase memory systems. Not 2X/year, but by orders of magnitude across new generations. This will enable a structurally new Al industry, including restarting scaling laws, holding entire repos in context windows & more. Our team has pioneered many core semiconductor technologies: the 1st CoWoS product, early HBM, the 1st silicon photonics CPO systems, the 1st high volume tunable VCSELs, the 1st processors to directly communicate using light & more. We’ve already sent data >10× farther than equally tiny electrical wires inside a chip package. Our next iteration is already taped out and targets world-record bandwidth density over relevant distances, read more: volantissemi.ai/news-insight…

What does RL add here compared to doing IK?

🤖 Should robots be generalists or specialists? At #IROS2026, @Ken_Goldberg & @andrea_bajcsy moderated one of the best debates in robotics. Matt Mason opened with a better question: “Should animals be humans?” 10 takeaways 🧵👇 1️⃣ @HarryXu12, for generalists: “Don’t build Excel. Build GPT.” We don’t know what we’ll need tomorrow, so buy generality just in case — and amortize it. The twist: GPT didn’t replace Excel. It learned to drive it. 2️⃣ @GeorgiaChal split “general” into 4 axes: Body. Scene. Contact. Task. “Breadth and reliability are measured per axis. One does not imply the other.” Her prediction: “The next task is the test.” 3️⃣ The most honest number of the day, from Georgia’s slides: A general model calling robot tools. Coarse block placement: 19/20 ✅ Precision insertion: 2/20 ❌ “Coarse pick and place works. Physical composition and precise contact not yet.” Tool use ≠ dexterity. 4️⃣ @KaterinaFragiadaki’s architecture: The generalist writes the curriculum. The specialist does the reps. Video → sim → RL on subgoals → specialist policies. No teleop. At deployment: “Fast specialists for the common case. Flexible generalists for the long tail.” 5️⃣ Specialist bench: “Fixture and conveyor are part of the intelligence.” A fixture is a prior. A conveyor is a scheduler. Strip them to look “general” and you’ve moved work onto a pricier learned controller. Optionality that hurts reliability isn’t intelligence. It’s cost. 6️⃣ A surgical robot doing one procedure 10,000 times learns exactly what can go wrong. “Generality is the enemy of certification.” From the floor: Are human drivers generalists or specialists? We certify generalists one task at a time. That’s what a license is. 7️⃣ Toshio Fukuda changed the register: “NON-AI matters.” Kizuki: help offered before it’s asked for. “Awareness comes from the body, not from the model.” Japan: ~29% aged 65+, ~570k care workers short by 2040. “If people live to 120, must our machines last that long?” 8️⃣ The floor fight: “A human is a generalist.” “We’re all human specialists.” The room accidentally re-derived pretrain-then-finetune. Veterans argued from what has shipped. Newcomers argued from what has scaled. Both are data. 9️⃣ The title asked for a side. The slides drew a stack: General planning + recovery on top. Fast, certifiable specialists below. A verifier is a fixture made of code. 🔟 The real question: How thick can the middle get before the body must specialize anyway? Generalist or specialist? Maybe the winning architecture is neither. It’s a stack. #IROS2026 #Robotics #PhysicalAI #RobotLearning #WorldModels

Instead of conferences, we should do more challenges. Submit papers that solve a big challenge, with a shared eval for every submission. Basically how challenges and prizes have always worked. We can still push research, but with less slop and a clear winner. Maybe the challenges are community-pooled, and the evals are too. Then the best paper basically boils down to winning on everyone else’s evals lol. And if the community can’t even come up or agree broadly with a particular challenge that someone solves, maybe the problem shouldn’t be solved in the first place, like who cares. This will eventually fight the goal-oriented vs idea-oriented research debate, bringing them to the same house. We can move faster than ever.

If you are in the data business. You should know whats on the horizon - Post Training Data from Deployments (failure, novel OOD data) - Scaling Data from Simulation (once Real2Sim is 1:1 fidelity) - The flywheel between the two, creates a new type data mixture that might fade or decrease pre-training data The great data companies WILL move into newer forms of data, eventually.

We use URR to unify both data and control. Our RL System 0 tracks URR directly in the world frame, letting us assign different objectives and priorities to different parts of the body. Teleop, off-robot data, and Dyna-2 models all produce URR. That gives us a clean boundary between task-level intent and low-level execution. A nice outcome is that we can avoid robot-specific retargeting in our teleop and off-robot data pipelines.