T
traeai
Sign in

概念

什么是 VLA

也叫:vision-language-action

视觉-语言-动作模型,具身智能核心技术路线。

为什么现在值得关注?

最近变化

2026-06-08 · 2026年AI行业估值跳涨与价值重估并存,创业者需谨慎选择赛道。

VLA 被反复提及时,通常意味着它正在影响产品路线、开发者工作流或 AI 产业判断。这个页面把分散材料合并成一个可持续更新的观察入口。

📰 VLA 最新动态

已收录 8 篇与「VLA」相关的 AI 资讯和分析。

刚刚,全球⾸个“事件级预测”具身智能世界模型来了!

ZiBianLiang Robotics launched WALL-WM, the world’s first event-level prediction embodied world model, replacing frame-based prediction with semantic events (e.g., 'grasp', 'place'), significantly improving cross-scenario generalization and action robustness.

入选理由:WALL-WM以语义事件(如抓取、抬升)为建模单元,替代传统固定时长动作块,使动作长度可变且更符合物理逻辑

FeaturedArticle#Embodied AI#World Model#VLA#Event Modeling#Robot Learning中文
“VLA和世界模型都不是终局,会有物理世界独有的模型” | 蚂蚁灵波沈宇军@AIGC2026

Shen Yujun believes that VLA and world models are not the endgame for embodied intelligence; a unique physical-world model will emerge. Robots lack data, so spatial perception must be built from sensor inputs.

入选理由:机器人行业缺乏物理世界数据,需构建专属物理模型

FeaturedArticle#AI#Robotics#Embodied Intelligence#VLA#World Model中文
CVPR 2026,英伟达特斯拉Waymo一块听中国公司讲物理AI

XPeng unveiled its Physical AI foundation model at CVPR 2026, achieving co-evolution of dense physical prediction and sparse human intent by fusing Gen-2 VLA with World Models. Leveraging X-World tech and self-developed chips, it reduced vehicle inference latency to 80ms, validating Autonomous Driving Scaling Laws.

入选理由:小鹏第二代VLA与世界模型协同进化,单版训练Token破4万亿,解决稀疏意图监督难题。

FeaturedArticle#Physical AI#World Model#VLA#XPeng#Autonomous Driving中文
一个GPT Plus会员的钱,够机器人跑一个月世界模型了

A GPT Plus Subscription Can Run a Robot's World Model for a Month

量子位2895 字 (约 12 分钟)
85

The world’s first implicit world model, Being-H-Flash, runs in end-device chips at 150 USD/month for a single robot—about a tenth of the cost of NVIDIA Cosmos and 70% cheaper than Pi0.5. It runs near-real-time at ~20 FPS on Orin NX and supports both NVIDIA and domestic platforms, reducing latency and connectivity risks and enabling large-scale deployment of world models.

入选理由:单台机器人月算力成本降至150元,实现端侧实时世界模型推理。

FeaturedArticle#embodied world model#implicit world model#end-side deployment#100TOPS#Being-H-Flash中文
165: 英伟达 GEAR 高深远:世界模型、自进化循环、DreamDojo

165: NVIDIA GEAR High Yuan: World Models, Self-Improving Cycles, DreamDojo

晚点聊 LateTalk1447 字 (约 6 分钟)
85

NVIDIA GEAR Lab introduces DreamDojo and DreamZero world models, advancing embodied intelligence.

入选理由:英伟达 GEAR 实验室发布 DreamDojo 世界模型,支持视频数据训练。

FeaturedPodcast#AI#Embodied Intelligence#World Model#Self-Improving Cycle#NVIDIA中文
机器人的终局:英伟达 Jim Fan 宣告 VLA 时代结束,WAM 登场

机器人的终局:英伟达 Jim Fan 宣告 VLA 时代结束,WAM 登场

宝玉的分享2719 字 (约 11 分钟)
85

英伟达Jim Fan宣布VLA路线过时,提出新范式WAM,代表作为DreamZero,预计2040年实现机器人自主设计制造,置信度95%。

入选理由:VLA路线过时,新范式WAM登场

FeaturedArticle#机器人#英伟达#WAM#DreamZero中文
95% sure robotics is fully solved by 2040 | NVIDIA's Jim Fan

95% sure robotics is fully solved by 2040 | NVIDIA's Jim Fan

Sequoia Capital252 字 (约 2 分钟)
72

NVIDIA科学家Jim Fan在Sequoia AI Ascent会议上提出,机器人技术仅剩3项关键突破,95%确信2040年前将实现通用具身智能的全面工程化落地。

入选理由:机器人技术瓶颈已收敛至感知-推理-执行闭环中的三个具体缺口:长程任务规划、零样本技能迁移、物理世界鲁棒交互。

FeaturedVideo#robotics#AI#NVIDIA#embodied AI#VLA中英混合

与「VLA」经常一起出现的 AI 术语。

💡 想追踪「VLA」的长期趋势?去 实体雷达 · VLA 查看详细分析和跨材料问答。

AI may generate inaccurate information. Please verify important content.