Unverified50% confidenceOpinionExact time
Krishna认为机器人模型应直接从视觉与传感器输入映射到动作，而非在中间嵌入LLM
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
LLM是机器人智能的错误地基？具身AI的路线之争
Related Claims
UnverifiedVLA将机器人的物理动作表示为动作token，在统一序列预测框架中端到端完成从感知到行动的推理，Google DeepMind的RT-2系列工作是这一范式的代表68% similarUnverifiedAndrew Ng's team is exploring the direction of embodied intelligence that combines large-model-driven agents with robotic hardware.67% similarUnverified川虎ChatGPT内置Agent智能体功能，允许模型调用外部工具执行多步推理任务66% similarUnverified智能体的许多能力实际上是由Agent所运行的框架在LLM外部实现的，而非LLM本身65% similarUnverified基于视觉语言模型（VLM）的屏幕理解使Agent能以语义层面而非像素层面理解界面，这一范式转变被业界称为认知自动化（Cognitive Automation）64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/507965API
curl https://kongchang.com/api/v1/knowledge/claims/507965MCP
get_claim(id=507965)