NVIDIA releases Cosmos 3, an omnimodal world model for Physical AI
One model to reason, generate, and simulate across text, images, video, audio, and robot actions.
Ranked #1 open-source for image generation, video generation, and robot policy.
A simple MPC controller for path tracking implemented in python.
Few resources actually teach it without assuming a PhD first.
This open-source project does exactly that:
An iterative MPC tracking controller (built with CVXPY and MuJoCo) that shows how to make a robot (or autonomous vehicle) follow a path and avoid obstacles in real time.
→ Autonomous logistics, robot navigation, vehicle control; there's a path-following algorithm behind all of it.
→ Most implementations are black boxes. This one shows you what's actually happening.
→ It uses convex optimization (no non-linear solver), and still handles real vehicle physics through iterative linearization.
What makes it practical is that you can run it with the mushr car model in MuJoCo, or headless without any physics simulation if you just want to understand the logic.
If you're making the jump from classical control to real-time optimizatio, this is a very good starting point.
📌 Repository: https://t.co/dzi784Vgmq
——-
Weekly robotics and AI insights.
Subscribe free: https://t.co/9Nm01QUcw3
Spotted at ICRA - magnetic levitation
Walking around the expo I came across this levitation demo. A steel ball suspended in mid-air, held up by an electromagnet balanced against gravity.
The physics make it a good teaching tool.
Magnetic force drops off with the square of distance, so the system is inherently unstable - any overshoot and the ball either falls away or snaps into the magnet. There’s no passive equilibrium.
Position is measured via the ball’s shadow, current is adjusted at 50Hz, and your job is to tune the PID controller to keep it stable.
Controller tuning is one of those things that’s hard to develop intuition for from theory alone.
A system that punishes bad gains instantly closes the feedback loop in a way that simulation doesn’t quite replicate.
The company is called Inteco…
Open source Text2Speech just hit a new level 🔥
Miso TTS 8B is a new model built for highly emotive, human-like conversations.
→ Local inference
→ Watermarked outputs
→ Voice cloning support
→ 8Bparameter architecture
100% Open Source.
GitHub 🔗: github. com/MisoLabsAI/MisoTTS
At #ICRA2026, certifiable optimization is having a moment. Our workshop paper with Avinash Subramanian, Connor Holmes, Tim Barfoot & Frederike Dümbgen brings chordal sparsity + Bayes trees to globally optimal factor-graph estimation in GTSAM. Blog post and arxiv link in the replies.
How do we get the workforce to keep up? Automation isn't a nice-to-have anymore – it’s the only way to meet the demand. Stay tuned to find out how Akeratos is solving the automation problems others walk away from. https://t.co/ZxE8HMMM70
Congrats for release, I am glad to help setting up the cosmos 3, which has two policies submissions on RoboArena (Cosmos3-Nano-Policy, T1, Score 1867, SD 122, runs 20; balsomsoc, T6, Score 1549, SD 33.9, runs 234)
The ranking is changing really rapidly, please stay tuned! :)
NVIDIA just made a major move in the humanoid robot race.
At GTC Taipei, Jensen Huang unveiled Isaac GR00T, an open humanoid robot reference platform designed to become the foundation for the next generation of Physical AI.
Key specs:
• Unitree H2 Plus humanoid body
• Jetson Thor AI computer powered by Blackwell architecture
• 75 degrees of freedom
• Advanced five-finger dexterous hands
• Stereo vision and real-world perception
• Built for seamless sim-to-real training
Why this matters:
→ AI models got their "ChatGPT moment"
→ Humanoid robots are now getting their "Android moment"
Instead of every company building everything from scratch, NVIDIA is creating a common hardware and software stack that researchers and startups can build on.
Stanford, ETH Zurich, and other leading robotics labs are already on board.
The goal isn't just smarter chatbots.
It's billions of AI-powered machines that can see, understand, and interact with the physical world.
The next trillion-dollar AI opportunity may not live in the cloud.
It may walk on two legs.
Let’s build an on-device voice agent by fine-tuning a small 1.5B parameter speech-to-speech model by @liquidai
Hands-on tutorial with video recording and code. Enjoy ⬇️
https://t.co/wCzhrYvjr3
Claude Opus 4.8 just dropped and it's a monster
• New ULTRACODE mode
• Dynamic Workflows (actually sick)
• Cheaper /fast mode
• Hints of Mythos
In this video I cover every change and show you how to take advantage of them
(and trust me, you want to be using workflows)
🎥 "Without PSL Datatrack, we wouldn't have been able to get to the stage we are at."
Charlie of Caffyn Engineering reflects on how #PSLDatatrack has supported the business through rapid growth! Watch the full video ▶️ https://t.co/arihPOxUsf
#PrecisionEngineering@mtdcnc