看影片(在新分頁開啟原站)連到 AI Engineer
摘要
Deepak Pathak 指出機器人學卡在硬體思維,缺乏通用大腦,並展示 Skild AI 的「全身體」模型如何透過模擬、人際影片與遠端運算元據訓練。他說明該模型能跨硬體執行插耳機、煮蛋等任務,並透過部署資料飛輪持續進化,甚至讓斷腿機器人重新學習行走。
Deepak Pathak argues robotics is stuck due to hardware focus and proposes an omni-bodied brain model that scales via simulation, human video, and deployment data flywheels.
摘要、重點與章節標題由語言模型整理,細節(誰說的、數字、先後)可能有誤;要引用請以原始內容為準。
重點
- 機器人學因過度關注硬體而停滯七十年。
- Skild AI 提出單一模型處理任何機器人與任務的架構。
- 透過模擬、人際影片與部署資料的飛輪實現快速進化。
章節
依話題轉折切分,標題由 AI 產生
- 00:00AI's progress and the robotics hype
- 01:02Robotics has been "almost here" for 70 years
- 01:27A 1960s block-copying robot
- 02:17Teleoperation in 1957
- 03:57Why robotics is stuck: the missing general brain
- 04:27Moravec's paradox
- 05:42There's no internet of robot data
- 07:22Skild's thesis: one brain, any robot, any task
- 08:22Teaser: many robots, one brain
- 09:31How to scale robot data: a career retrospective
- 12:01The three properties of robot data that matter
- 13:21Pre-training, post-training and the deployment flywheel
- 14:56Why laundry folding is actually easy
- 16:17The AirPods task
- 17:21Learning from human videos
- 18:36The robot's "dream": imagined training data
- 18:56Cooking omelets on $4,000 arms
- 20:41Truly end-to-end: camera in, motor power out
- 21:21Why stairs are harder than backflips
- 24:50Deployment: GPU assembly for NVIDIA
- 26:00Package delivery to the front door
- 26:25Safety: adapting when the robot breaks
提到的工具與公司
- NVIDIA
- GPT-3
適合誰看
對機器人學、AI 模型架構或自動化部署感興趣的開發者與研究者。
摘要依據
- 講者
- Deepak Pathak
- 依據
- 自動字幕
為什麼排在這裡
- 人氣
- 0.83
- 新鮮
- 0.96
在主題頁與搜尋結果裡,名次由相關、人氣、新鮮三個分數決定;這一頁沒有搜尋的關鍵字,所以沒有相關分數。排序怎麼算
相關內容
- One Brain, Any Body: Google DeepMind's Keerthana on Gemini Robotics 2, Cross-Embodiment & HumanoidsPodcast ・ The Cognitive Revolution ・ 1 小時 32 分
- EP 44.【AI年终特辑3】具身智能深度对话:从学术到产业,机器人的ChatGPT时刻来了吗?Podcast ・ OnBoard! ・ 1 小時 51 分
- 于是转身向具身走去|对话王家伟:24 岁的具身智能首席科学家Podcast ・ 十字路口Crossing ・ 1 小時 10 分
- Ep 70: Karol Hausman and Danny Driess (Physical Intelligence) Unpack the Most Recent Breakthroughs & Path to Generalist RobotsPodcast ・ Unsupervised Learning ・ 1 小時 10 分
- How Physical AI Learns Across Language, Video and Action — Ming-Yu LiuPodcast ・ Machine Learning Street Talk ・ 26 分鐘
- Redefining Robotics with Carolina ParadaPodcast ・ Google DeepMind: The Podcast ・ 45 分鐘
摘要由 AI 根據原文產生,可能有誤;完整內容請看原站。看影片(在新分頁開啟原站)
