Human life is finite, the impact of purposeful action can outlast a lifetime. Develop a strong sense of urgency, we have a lot to achieve, people.

What if motion planning, #simulation, and fleet management were already built in? In this video on YouTube, Brandon walks through building a palletizer with Viam and getting boxes moving in about 30 minutes: youtu.be/hg4s2FmWRZc

Self Sabotage. Proudly sponsored by Dario

The process of how world models reason: The major purpose is to have a model that understands the physical world. On top of this, we are using LLM to interpret words for input and output but the reasoning layer is completely done by the world model itself. The way further is either: Using a vision model (see objects and reason) Or directly running through hardware (sensors and world state) View on a wider screen, I'll push up the frontend to be better.

A very interesting question in this exploration is, How do we optimize inference for knowledge reasoning on the hardware architectures we have today? If we move toward world reasoning models, where systems continuously maintain and update representations of their environment, is the bottleneck really the model? Or is it the software and hardware stack around it? In robotics, we are increasingly running reasoning models on top of ROS and existing infrastructure. But does embodied intelligence eventually require a new runtime architecture designed around memory, planning, reasoning, and action from the ground up? Curious how others think about this. Do we need better models, better hardware, or a new systems architecture for world reasoning?

while building the agentic mode in SimArena, I kept thinking about the UX problem most GenAI platforms have: you fire off a prompt and wait minutes for the entire scene to load before you can interact with anything I worked with @itsdamnvinci this week to make SimArena's workflow smoother. Instead of blocking operations, jobs run concurrently. scene objects stream in as lightweight placeholder boxes, then swap to final @NVIDIARobotics SimReady assets now you can keep working while your scene renders. no more waiting

VLA-JEPA just dropped in LeRobot 🤖 What makes this model special is that it does not just learn what action to take from a given observation, it also leverages a JEPA world model to learn action-relevant dynamics. During training, the VLA leverages V-JEPA2 by conditioning its predictor. This clever trick adds a world modeling objective to the training, which also allows pretraining on human videos. At inference, the world model is dropped entirely, keeping only a standard VLA architecture: Qwen backbone and action head. The demo here was only fine-tuned on 13 examples, showing great pretraining capability and running in real time on @NVIDIARobotics DGX Spark! VLA-JEPA is the first world model to be ported to LeRobot, and I feel like it won't be the last 🚀 @Thom_Wolf @ClementDelangue

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Its capabilities exceed those of any model we’ve ever made generally available.

How world models turn raw vision into structured spatial representations

QVAC by @tether is live on peaq’s robotic.sh Private, on-machine AI inference for robots and machines Run it on the machine itself, or on one right next to it. The data stays private. → peaqOS provides identity and coordination → @qvac by Tether provides private, on-machine inference robotic.sh/services/qvac

Diffusion reasoning models are going to be discussed a lot in the coming months, but in a very different way from how we've approached them so far.

Got a very beautiful rejection mail from YC. Let's get this done regardless..

Big dreams require extraordinary dedication and sacrifices.

There are some things you should learn to see before everyone else does. It may sound crazy, but when you see them, act on them.

Founders are getting accustomed to the cost real quick.