Tesla, Figure and the Fight Over Humanoid Robot Hands | Scott Walter, RoboStrategy
Episode notes and transcript: https://corematter.substack.com/p/humanoid-robot-hand-design-tradeoffs Subscribe for engineering conversations about robot hardware and deployment. Tesla, Figure and 1X have taken different paths through humanoid robot hand design. Scott Walter explains why tendon routing, direct-drive motors, tactile sensing and task fit keep producing radically different answers. The conversation begins with the designs that made the debate visible, then moves to three useful extremes. Allonic braids the finger structure. Daxo pushes to as many as 120 tendons. Tacta treats the hand, actuation, sensing and data collection as one system designed for cobots and industrial arms, outside the humanoid constraint set. Scott and Michelle also leave the hand debate for a proposed humanoid decathlon, the strange running forms that emerge when robots stop copying people, data-center cable insertion, and a broader question: if humans have general-purpose bodies and specialized professions, why should robots converge on one universal design? Recorded remotely in August 2026. Guest: Scott Walter, PhD, Robotics Research Diligence Director, RoboStrategy, Inc. Host: Michelle Sun, Core Matter Core Matter, independent research on the Physical AI stack and supply chain: https://corematter.substack.com Chapters 00:00 Cold open: the labor prize and the 5-year forecast 00:24 The Fermi paradox of robotic hands 01:18 Tesla, Figure and the tendon debate 05:10 Bowden tubes and routing tendons through a wrist 11:15 Joints, actuators and real degrees of control 14:38 Why the pinky matters more than it looks 17:23 Humanoid Games as real-world robotics experiments 21:57 Scott's humanoid decathlon challenge 26:28 Tesla's hand iterations and abandoned designs 27:55 Wuji and Sharpa Wave put motors in the fingers 38:29 What 3 extreme hand designs can teach us 38:53 Allonic's braided hand and portfolio disclosure 51:15 Daxo's maximalist hand with up to 120 tendons 01:02:36 Tacta measures how workers use their fingers 01:05:18 A robotic-hand system designed outside the humanoid 01:13:12 Data-center cables as a tactile manipulation task 01:16:22 General-purpose bodies and specialized work 01:19:18 Why this 5-year robotics forecast may be different
Read Video · 文字起こしと深掘り分析
このエピソードには全文文字起こし + AI インサイトがあります
無料アカウント · カード不要 · 登録で150クレジット獲得、このエピソードのアンロックに十分
- 📄 タイムスタンプ付き全文文字起こし
- ✨ AI 要約・キーワード・マインドマップ
- 💡 重要ポイントと名言
エピソードのタイムライン
The Fermi paradox of robotic hands: why dexterous hands exist in labs but never reach deployment
- Robotic hands have long been the unsolved bottleneck of robotics because they must combine dexterity, controllability, and robustness simultaneously, and machine learning in the 2010s finally made control possible while robustness and cost remained unsolved.
- The Shadow Hand proved that a tendon-driven hand connected to good ML algorithms could achieve remarkable dexterity, but it failed on robustness and cost, which is why hands stayed in labs rather than reaching the field.
- Elon Musk's announcement of a 22-DOF hand for Optimus triggered a Cambrian explosion of robotic hand designs, splitting the field into tendon-based and non-tendon-based (direct drive) camps.
Brett Adcock's claim that tendon-based hands are an engineering dead end ignites a fierce debate
- Brett Adcock publicly stated that Figure's first-generation tendon-based hand was the biggest engineering mistake he ever made and called tendons an engineering dead end, provoking a spirited industry argument.
- Tendon proponents responded that it is a skills issue rather than a design flaw, while others confirmed from their own experience that tendons turned out to be really hard to make work.
- Even within the tendon camp there are fierce internal disagreements about the best approach, and 1X's 22-DOF hand with roughly 44 tendons shows how complicated these designs become.
Routing tendons through the wrist is the hardest mechanical problem in tendon-driven hands
- The biggest challenge is not moving fingers but routing tendons through the wrist, because the wrist adds a layer of complexity that makes engineering solutions very difficult.
- In theory tendons could pass through the wrist's center of rotation to keep length constant, but the Pauli principle means two particles cannot occupy the same point in space, so there is never enough room.
- Bowden tubes, borrowed from bicycle brake cables, redirect tendons and maintain constant length, but they add friction, take up space, and can snap, which is why Tesla's first hand used them and later abandoned them.
主要な概念
- tendon-based hand— The central debate of the episode: whether tendon-driven robotic hands are the right architecture or an engineering dead end.
- direct drive hand— The rival architecture where actuators sit directly at each finger joint, championed by Figure and others.
- degrees of freedom— The recurring metric for hand dexterity, with human hands estimated at roughly 20-22 DOF.
注目の名言
That is the biggest pie we've ever seen, ever, trying to automate labor.
🔥— Frames humanoid robotics as the largest economic opportunity in history, far beyond any previous automation wave.
And he called it the biggest engineering mistake he'd ever made.
🤯— A CEO publicly branding his own tendon-hand effort a catastrophic error is a rare, candid admission that reframes the entire architecture debate.
実行可能なテイクアウェイ
⚙️Engineering Tradeoffs
Every hand architecture is a compromise: tendons keep fingers light but add routing complexity, while direct drive simplifies control but adds mass and heat at the fingers.
This week, pick one robotic hand design and write a one-page pros/cons table listing its actuator placement, mass distribution, and failure modes.
Scaling laws break intuition: motors lose torque density as they shrink, which is why remote actuation wins at human scale but direct drive may win at micro scale.
Sketch a torque-density vs. scale curve for a motor you can find specs for, and mark where tendon vs. direct drive crosses over.
🔬Learning from Extremes
Designing for extreme scales or constraints teaches unexpected lessons, like the 1.3-meter hand or the 120-tendon Daxo hand.
Pick one extreme design from the episode and write down three lessons it teaches that apply to normal-scale hands.
You are a humanoid, so you can run hand experiments on yourself by taping fingers or wearing thick gloves.
Spend one hour this week doing daily tasks with your pinky taped to your ring finger and note which tasks become hardest.
文字起こしとAIインサイトは自動生成されたものであり、誤りが含まれる場合があります。認識精度は音質や話者の発話の明瞭さに左右されます——内容に不就がある場合は、元の音声をご確認ください。
すぐに読めるエピソードと動画
ポッドキャスト

How I built a machine that feeds millions | Felix Brooks-church
TED Talks Daily
2026年7月22日17:51EN
The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang | BG2
BG2Pod with Brad Gerstner and Bill Gurley
2026年6月11日1:20:47EN
87 | On the joys of being a young scientist
Night Science
2026年7月13日38:07EN
#4「見た目を草間彌生にした方がいい」
朝井リョウ・加藤千恵 信頼できない語り手
2026年2月27日46:24JA
#58 Hexenhammer und Eulenspiegel - Die Bestseller des frühen Buchdrucks
Wer wir sind und warum das nicht klappte ...
2026年5月20日1:11:23DE
la teorías de las parcelas 2
Más de lo Mismo
2025年11月2日36:26ES
動画

No Priors Ep. 80 | With Andrej Karpathy from OpenAI and Tesla
No Priors: AI, Machine Learning, Tech, & Startups
2024年9月5日44:15EN
Steve Jobs' 2005 Stanford Commencement Address (with intro by President John Hennessy)
Stanford
2008年5月14日22:10EN
Finally, some truth about loop engineering
NeetCode
2026年7月28日17:18EN
INDIA’S GOT LATENT S2 EP4 ft. Karan Aujla, Tanmay Bhat, Gurleen Pannu, Rahul Dua
Samay Raina
2026年8月2日53:54HI
#32 Die Schlacht von Worringen - Der Freiheitskampf der Kölner
99 mal Geschichte
2025年11月12日53:29DE