36Kr AI surveys 32 world models for physical AI
36Kr AI compiled 32 world models and research systems for physical AI, covering robotics, autonomous driving, simulation, and policy training. It describes world models as learned internal simulators that predict environmental changes from observations and proposed actions, with systems from Nvidia, Baidu, Unitree, Google DeepMind, Meta, Waymo, Wayve, Xiaomi, and others. The report also notes limits involving deformation, disappearing objects, collision and sliding inaccuracies, accumulated prediction errors, generation speed, and computational cost.
The survey maps several major routes for using world models in physical AI, but many capability claims come from company disclosures or research papers and require independent validation.
让机器人拿起一只杯子,它要解决的问题远比识别物体多:杯子在哪里,机械臂从哪个方向伸过去,夹爪什么时候合拢,移动过程中会不会碰到桌沿。无人车经过路口时,也要判断:如果现在加速,旁边的车和行人接下来会怎样行动。
这类需要在现实环境中感知、决策和行动的AI,通常被称为物理AI。人形机器人、自动驾驶汽车、机械臂和无人机,都是它的应用载体。重量、惯性、摩擦和碰撞,会直接影响它们能否把一项任务完成。
世界模型是帮助AI预测环境变化的一类模型,可以将其理解为系统内部一个通过数据学习的模拟器。它接收当前观察,结合动作等条件,推演接下来可能发生什么,为规划、训练和评测提供依据。
以拿杯子为例,模型可以尝试预测:机械臂沿某条轨迹移动,能否接近杯子;夹爪抬起后,杯子会不会随之移动,是否可能滑落。把候选动作对应的未来结果交给规划系统比较,就有机会在真实执行前发现不合适的方案。
#Nvidia#Baidu#Unitree#Google DeepMind#Meta#World Labs#Waymo#Wayve#Xiaomi#Tencent#AgiBot#ByteDance