Very cool work
11.03.2026 20:13
π 0
π 0
π¬ 0
π 0
Very cool work
Why do they have a bad reputation?
1. Starship robots seem to be moving faster compared to a few years ago. What changes helped achieve that?
2. How much processing is on board versus the cloud?
1. What are the major roadblocks to using more robots in construction?
2. How autonomous are the robots you deploy (percentage, maybe?) ?
Yup, feedback is a critical part of having imperfect processes reach their desired end. That is, closed loop and open loop are very different.
nice blog post about a humanoid robotics startup failure: ruixu.us/posts/six-th...
For all robotics or for robotics at Tesla?
Neat, do you describe the setup somewhere?
It HAS to be either autobotting or decepticonning
Those kp values look high, try 1000, maybe even 100.
CC vim mode is off to me
What do you like about diffusion models?
If the fingers are set to try to snap together, maybe reduce the gains, and try to make the set point equal the width of the object.
Try increasing the number of iterations for the dynamics solver. If that fails then possibly reduce the stiffness of the contact model, though you'll get more intersections between bodies.
What local model are you using?
Perhaps that's another good reason to not frame the number of citations as a measure of recognition.
Very meta, too! Neuro-inspired observation of neural phenomena.
Did you bolt the arm straight into the wood?
Fair, but does that improvement need to occur through the journal system? There's an argument for journals enabling expert feedback outside of buddy networks, OTOH.
"Imagine landing your dream job, only to discover that half your workday will be spent not on the work itself but on cold-calling strangers to beg for money."
Do these tear up the grass/soil any?
Preach
Also, a list of my favorite robotics papers:
github.com/philfung/awe...
Snippings of recorded talks
Last weekβs #AI4Robotics workshop was fab & fabulously full π If you missed out we've good news -
π½οΈ All 15 talk recordings are now online π! tinyurl.com/4ns5apvd
Catch up on cutting-edge work in #robot perception, action & autonomy - from #SLAM & control to computer vision & large-scale learning!
Loving Justin Carpentier's talk right now. Nice workshop!
With that soundtrack it feels like it could be a scene from the animatrix.
Collaborative news curation as a service, demonstrating the value of social media based on protocols, not platforms and a βplaybook for killing information voids.β
AI tornado!
Ideas: AI hoover, AI funnel, AI suck up (vs trickle down?).
Schematic showing how changes in prompts map to changes in a modelβs latent space. Left panel, titled βPrompt Spaceβ: a beige, wavy sheet with thin contour lines and many black dots. Two dots are labeled x (lower middle, with an arrow pointing to it) and xβ² (upper right). A short dotted segment between them is labeled Ξ΄. Right panel, titled βLatent Spaceβ: a square grid with xβy axes (a right-angle mark at the origin) and the label βα΅. Two points are shown: z (lower) and zβ² (upper right). A short dotted segment between them is labeled Ξ΅. Two long curved arrows map from the sheet on the left to the grid on the right: the upper arrow labeled Llm lands on zβ²; the lower arrow labeled SipIT lands on z. Centered below: Ξ΄ > 0 β Ξ΅ > 0.
your embeddings are not safe!
every prompt directly maps to its embedding and back. theyβre isomorphic
SipIt is a linear time algorithm for quickly and efficiently extracting input text from embeddings
www.arxiv.org/abs/2510.15511