具身智能系统 / EMBODIED AI SYSTEMS
VLA/VLM 机器人，从仿真数据到真实运行
VLA/VLM Robotics, from simulation data to robot runtime
我构建机器人智能背后的系统栈：多模态感知、策略学习、机器人大小脑架构、Real2Sim 数据闭环，以及可观测的运行时基础设施。
I build the stack behind robot intelligence: multimodal perception, policy learning, robot big brain / small brain architecture, Real2Sim data loops, and observable runtime infrastructure.
$boot vla-vlm stack --mode embodied
>perception=vlm policy=vla sim=real2sim
>brain=planner+controller data=observable
>status=online latency=adaptive loop=closed
机器人脑图 / ROBOT BRAIN MAP
技术栈总结
感知、策略、仿真数据与运行时基础设施不是分散模块，而是一套持续反馈的机器人智能系统。
Perception, policy, simulation data, and runtime infrastructure are designed as one feedback system.
感知 VLM
场景理解、多模态 grounding、物体状态和可供性信号。
Scene understanding, multimodal grounding, object state, affordance signals.
策略 VLA
将语言目标映射为可执行机器人动作，并接入评估闭环。
Language goals mapped into executable robot actions and evaluation loops.
机器人大小脑
高层认知连接小脑控制、运行反馈、记忆与工具系统。
High-level cognition connected to small-brain control, runtime feedback, and memory.
仿真数据
合成场景、Real2Sim 资产、数据集 QA 和 Sim2Real 验证。
Synthetic scenes, Real2Sim assets, dataset QA, and Sim2Real validation.
