A tech blog focused on Embodied AI, World Models, and Sim-to-Real transfer. From algorithm research to engineering practice.
Author: MSc at Northwestern Polytechnical University · Robot AI Engineer
From sensors to actuators: the complete pipeline of world models in robotic systems — perception fusion, latent dynamics prediction, policy learning, …
From RSSM to imagination, understand how Dreamer trains policies inside an internal world.
Wrap-up: split the config into architecture vs. training tables, compress RSSM into four core formulas, present the code↔math↔semantics mapping table, …
Future simulation without observation: the full imagine loop, the essential difference between Observe and Imagine, why imagination cannot be …
The core of training the world model: the complete Observe-phase data flow, the gradient routing of dyn/rep KLs, free_nats as a loss floor, and how …
Trace the real deterministic dynamics: how _core() does one-step transition, why deter is 8192, the block-wise parameterization of Block GRU, and what …
Translate the code into math with correct time indexing; understand the prior/posterior pair, straight-through categorical sampling, and how unimix …
Series opener: where RSSM fits in DreamerV3, why the stochastic state is not 'an integer then one-hot', and how real observations enter RSSM (the …
The World Model Hype Has Been Exaggerated: A Technical Analysis - WorldSense Tech Blog
DreamerV3 GPU Infrastructure: Cloud vs Self-Built Cost Analysis - WorldSense Tech Blog
AI Is Looking for a Physical Shell: From Smart Glasses to Wearable Agents - WorldSense Tech Blog
Isaac Lab Installation Guide: From Zero to Running on AutoDL - WorldSense Tech Blog
Isaac Lab: From DreamerV3 to Industrial-Scale Robot RL Training - WorldSense Tech Blog
The Data Challenge in Robotics: Where Does Robot Learning Data Come From? - WorldSense Tech Blog
When World Models Meet Transformers: From RSSM to Large-Scale Sequence Modeling - WorldSense Tech Blog
DreamerV3 Training Tips: Lessons from Real-World Debugging - WorldSense Tech Blog
Four Paradigms of World Model Representations: A Comparative Analysis - WorldSense Tech Blog
MuJoCo vs Isaac Sim: How to Choose the Right Robot Simulation Platform - WorldSense Tech Blog
Domain Randomization: The Bridge from Simulation to Reality - WorldSense Tech Blog
TD-MPC: How World Models Enable Robot Control - WorldSense Tech Blog
World Models as Synthetic Data Engines for VLA Training - WorldSense Tech Blog
Is Sim-to-Real Too Hard? World Model-Driven Adaptive Transfer Methods - WorldSense Tech Blog
Building a World Model Lab from Scratch: A MuJoCo + DreamerV3 Practical Guide - WorldSense Tech Notes
ABot-World-0: 24-Hour Stable Inference from an Interactive World Model - WorldSense Tech Blog
VLA vs World Models: Which Will Prevail? - WorldSense Tech Notes
Embodied AI and RL: Career Prospects and Salary Guide in 2026 - WorldSense Tech Blog
World Models: 8 Years and the Same Bottleneck - WorldSense Tech Blog
World Models in 2026: Where Are the Real Opportunities? - WorldSense Tech Blog
Is World Model a Good Research Direction? An Engineer's Honest Assessment - WorldSense Tech Blog
How to Get Started with Reinforcement Learning: A Practical Guide - WorldSense Tech Blog
Embodied AI in 2026: What Breakthroughs Can We Expect? - WorldSense Tech Blog
Deep Dive into RSSM: The Core Engine of World Models - WorldSense Tech Notes
What Is a Robot World Model? An Engineer's Deep Dive - WorldSense Tech Notes
Can You Break Into Embodied AI Without a PhD? - WorldSense Tech Notes