TL;DR: Articulated 3D assets (doors that open, drawers that slide, laptops that fold) are essential for robot simulation but painful to build by hand — ~40 minutes per object with dedicated annotat...
TL;DR: Zero-shot reward tricks (like TOPReward) are elegant but limited. ROBOMETER asks: what happens if you actually train a reward model at scale? Built on Qwen3-VL-4B with three lightweight MLP ...
TL;DR: Need a reward signal for robot learning but don’t want to train a reward model? Prompt a VLM with “this trajectory completes the task: True or False?” and use the log-probability of the “Tru...
TL;DR: A pre-trained world model is only useful if it transfers to real-world tasks. Cosmos-Transfer translates simulator outputs into photorealistic video via ControlNet-style conditioning on edge...
TL;DR: Cosmos builds world models in two paradigms. Diffusion models generate video by iteratively denoising continuous latent tokens — flow matching (velocity prediction) replaces the earlier EDM ...
TL;DR: Physical AI — robots, self-driving cars, humanoids — can’t scale by training in the real world alone. A World Foundation Model (WFM) is a learned video simulator: given past frames and a per...
TL;DR: MCP lets Claude (and any compliant AI host) call your Django REST API as tools. This is a walkthrough of building a production MCP server on top of DRF — OAuth 2.1, discovery endpoints, ngin...
TL;DR: I built a classifieds app for Hot Wheels collectors. Validated the problem, ran pilots, launched — and it died quietly from lack of critical mass. This is about doing everything “right” and ...
Review of all the books I read in 2025.
TL;DR: Deep learning models are black boxes – powerful but opaque. LIME (Local Interpretable Model-agnostic Explanations) explains individual predictions of any model by fitting a simple interpreta...