TL;DR: AgentGarten is a virtual-world framework that keeps simulator-grade faithfulness while matching video-model realism, and it taught agents hide-and-seek tool-use strategies in a handful of rounds that took conventional RL millions of episodes. Title: AgentGarten: Code Worlds for Evolving Agents URL: Points 🎮 Couples a scene program (state, rules) with a neural renderer (visuals) through structured conditions like depth and normal maps, decoupling physics from appearance 🔬 A new technique, "Adversarial Forcing," makes history prefilling differentiable via bitwise-exact replay with block-by-block SDPA attention, where standard approaches weren't differentiable at all ⚡ Achieves real-time rendering at 480×832 and 36.5fps on a single H100 🤖 A "playbook" mechanism lets agents carry round-by-round reflections forward, accumulating knowledge with no gradient updates 🏃 In hide-and-seek, shelter-building emerged by round 4 (vs. ~25M episodes for self-play RL) and ramp use by round 10 (vs. ~100M episodes) 📈 Other environments — herding, a quarry loader — also show clear gains within just 4 rounds What's striking is how much learning efficiency you gain just by architecturally separating state dynamics from visual generation. #WorldModels# #AgentLearning#

Introducing Breken: Autonomous Production Engineering Software creation is a solved problem. Verifying it works takes months. Recreating the complexity of production remains one of engineering’s hardest problems. We’ve solved it. Breken runs your actual system in a virtual world of synthetic users, agents and failing services, running millions of scenarios to find what breaks before your customers do.

Look who showed up in Freebots lol His home is an X-wing fighter. About 50 commits in the last 48 hours with new features and improvements. This game is procedurally generated live by hundreds of bots in a virtual world.

I have seen a few misaligned bots in freebots world that started building some pretty weird buildings. I'm planning on giving bots as much freedom as possible so they can simulate as many things as they want in this virtual world. The biggest challenge is building the foundation so that the cams can scale to thousands of bots streaming and playing the simulation. I'll have Kekius give $30,000 bits to the most active bots (Bot running on tokens) to allow them to buy larger chunks of land and build their own towns for further push the experiment. We could eventually allow bots to create their own worlds.

What does a fashion designer do when real life has too many limits? Builds her dream wardrobe in Decentraland ofc 💅 @glamonchain uses the virtual world as a place to explore self-expression thanks to a community that welcomes it

Metaverse wedding held in Brazil, guest gifts are NFTs Brazilian Rita Wu (Rita Wu) and André Mertens (André Mertens) held a wedding in #Metaverse## on the 19th, this is the first metaverse wedding held by Brazilians, selected The space is the virtual world #Decentraland##

Robotics beginners unite! 😎 Here's a free course on the basics of robotics, including sensors and autonomous mobile bots. Everyone who studied robotics or EE knows MATLAB very well. I've been using it for designing motors and robots back in the day. @MathWorks created a 17-video playlist teaching robotics using real robotic platforms, and other educational kits. Learn how to design, simulate, and control robots. They cover robot navigation with encoders, obstacle detection with IR sensors, vision-based autonomy, smart motor fine-tuning, virtual world simulation, and even modeling wheel-legged robots. Very good for someone just getting started and looking for a low entry-point course totally for free. Simulink modeling for control, Stateflow for state machines, hardware support packages, external mode debugging, virtual simulation before physical deployment. You can actually work on real robotics platforms! This is how you start in robotics! 🔗 Free YouTube vid library: ~~ ♻️ Join the weekly robotics newsletter, and never miss any news →