Frank Dellaert

My Annual Reviews article on Factor Graphs in Robotics is finally out with a publicly accessible link: annualreviews.org/eprint/85PQDQY…

@_albertorod_ The 360 rotations are super cool :-)

Another key departure from human morphology is the infinite rotation of the actuators by removing all cables across joints. Eliminating all cables across joints gets rid of one of the key hardware failures and gives Atlas its unique superhuman mobility. Atlas can move backwards simply by inverting its legs and flipping its torso, rather than wasting steps and time turning around. One of my favorite things is seeing Atlas stand up. It makes it clear we have not built Atlas to be a human replica. To me it looks like it belongs in the Star Wars or Transformers universes. Meant to be intentional, effective, and useful.

Here is my take on one of the questions that I’ve heard most often recently: why humanoid? The thing that drives the current bet on humanoids is the perceived potential value of a generalist solution to physical work. Outside of some applications that have very large and very stable volumes, for all other physical work, automation becomes unfeasibly expensive if we have to specialize hardware, or models or deployment strategies.

@_albertorod_ If you’re only going to operate on flat ground, however, like on a factory floor, wouldn’t a heavy wheeled battery near the bottom benefit both stability and longevity? Sure, balance is not as hard as it used to be, but it still requires actuation effort all the time.

@aerospeedsta This type of continuous time representation has been explored for a while now, especially in SLAM. I would recommend downloading the paper which has a lot of references to earlier work led by Tim and many others.

@fdellaert I'm not sure exactly if your post fully or in any way reflects my ideation here, but I really found it uncanny that I thought about non uniform time intervals and how it could be resolved hierarchically in time and state in GPs and here I saw a similar idea being discussed.

@jbhuang0604 Nice! I learned a bunch :-)

The 60-Year Hunt for AI's Most Important Function I was trying to understand how SwiGLU works, but I couldn’t find an explanation that clicked for me. So I made this video to explain it from first principles. Check it out: youtu.be/JRaPNrpsQ9s

The only correct answer when a VC asks: "What's your moat?"

A haiku, after receiving some heavily AI-assisted class-work this Spring semester :-) Code is fungible! Is language, too? Humanity, forever changed.

@andrewgwils Code is fungible! Is language, too? Humanity, forever changed.

There's something refreshing about reading a pre-2023 paper and being absolutely certain that it was all human generated. As much as I will do research on LLMs, I will never use them to write a word of my papers.

The brains behind Mujoco ! (with Emo Todorov)

Co-written with @rbansal22, and we'll hear more from him next week ! :-)

Link to EqF blog post: gtsam.org/2026/04/28/equ…

New GTSAM blog post on a new capability: the Equivariant Filter (EqF). This one is geekier than last week’s STAG post, but if you're intrigued by equivariance, manifolds, or symmetry-aware estimation, this post provides a gentle intro to the EqF. teaser image below, link in reply.

@bilawalsidhu Wow. Pyramids paper link handy?

The same satellite tech can track a convoy through dense cloud cover at night, flag a sinkhole weeks before it opens, catch a village built on a collapsing slope, and even map the inside of a pyramid from orbit. It's called Synthetic Aperture Radar (SAR), and a lot of the data is free. In this video I cover the physics, the commercial arms race, and the defense constellation that will track every moving vehicle on Earth by the 2030s. 0:00 - A convoy no camera can see 0:56 - What is Synthetic Aperture Radar? 2:02 - InSAR: sub-millimeter change detection 3:07 - They scanned inside a pyramid from space 4:20 - These cities are sinking faster than sea levels rise 5:00 - North American bridges are the worst on Earth 6:09 - An AI caught a village built on a landslide 7:50 - The commercial SAR arms race 9:01 - GMTI: tracking convoys in the dark 11:26 - NISAR: the billion-dollar SAR satellite 12:45 - The other side of this all Full video on YT on Bilawal Sidhu channel. Enjoy.

@IlirAliu_ Wow, this is amazing! Do we still need rosbags after this? Ànd, will @foxglove support it ? :-)

Open-Source Multi-Sensor Data Platform for Neural 3D Reconstruction and Physical AI [📍github] It handles cameras, LiDAR, radar, poses, calibrations & labels in one clean format. No more messy custom parsers. • Super efficient (non-redundant storage) • New .itar single-file format with lightning-fast random access • Streams straight from S3/GCS/Azure – perfect for huge datasets • Built-in converters for Waymo, ScanNet++ & more • Already powers NVIDIA NuRec, 3DGRUT & gsplat Saw this at @jmartinezesturo, thanks for sharing! Easy to try: • pip install nvidia-ncore • GitHub: github.com/NVIDIA/ncore • Docs & project page: research.nvidia.com/labs/sil/proje… NCore slashes data wrangling time, cuts storage waste, and makes large-scale neural 3D training faster and simpler than ever. A real standard for physical AI. ——- Weekly robotics and AI insights. Subscribe free: 22astronauts.com

@AlexTensor KF is fixed-lag smoother with lag=1, linear dynamics, linear measurements, so a special case. Other extreme is large lag, high-dimensional states, learned nonlinear dynamics, CRF-style learned perception.

@fdellaert It sounds similar to Kalman filtering. How is it different?

That’s exactly the same discussion I had with Tim on LinkedIn :-) Answer: In STEAP we treated it as one graph, but I’ve come around to the separation principle :-) and I’m totally fine to run two different optimizations. At the end of the post I even talk about using MPPI for the optimization/search process on the right/future.

@fdellaert And then that's more in line with end-to-end approaches, but still has the nice undirected graph advantages for computation of a factor graph? 5/5

@ben_sdl I don't have the answer, but al least have "world models" in the title :-)

@fdellaert Looking forward to reading it ! Are going to cover also what factor graphs can do in the era of world models ?

New blog post coming soon. Teaser image courtesy of GPT Image 2.0 :-)