Models & Algorithms

Core embodied AI models and algorithms, from classic imitation-learning policies to VLAs, world models, LLM-based planning, and the open-source ecosystem around them.

Reading Path

• If you want the big picture first, start with Model Roadmap.
• If you want the main paradigms of robot foundation models, read Foundation Models for Robotics.
• If you want the VLA mainline from RT-1 to OpenVLA, pi0, and RDT, read VLA Models.
• If you want to understand why ACT still matters and how it bridges chunked policies and later action modeling, read ACT Model.
• If you care more about open-source reproduction and model selection, read Open-Source Model Summary and Datasets & Benchmarks.

Contents

• Model Roadmap — A timeline + paradigm tree of robot model evolution
• Foundation Models for Robotics — Three paradigms: LLM planning, VLM fine-tuning, purpose-built robot models
• VLA Models — RT-1, RT-2, Octo, OpenVLA, pi0, RDT
• ACT Model — Action Chunking with Transformers as a key bridge model
• World Models & Video Generation — Dreamer, UniSim, Genie, Cosmos, Genesis
• LLM-Driven Robotics — SayCan, Code as Policies, VoxPoser
• Open-Source Model Summary — Model catalog, frameworks, and quick starts
• Datasets & Benchmarks — Open X-Embodiment, DROID, LIBERO, RLDS

---

评论