Cover Image for Robotics & WM Reading Club 30: What Scale Doesn't Fix: Representation & Architecture in Robot Manipulation+Omni-Reason Video Model. SF 9/26

Presented by

🤖 Saturday Reading Club on Robotics & World Models for AI Researchers in SF
Hosts: Junfan Zhu, Aurora Feng
discord.gg/WH7DrTHRXK

Hosted By

138 Went

Robotics & WM Reading Club 30: What Scale Doesn't Fix: Representation & Architecture in Robot Manipulation+Omni-Reason Video Model. SF 9/26

Name: Robotics & WM Reading Club 30: What Scale Doesn't Fix: Representation & Architecture in Robot Manipulation+Omni-Reason Video Model. SF 9/26
Start: 2026-09-26T14:00:00.000-07:00
End: 2026-09-26T17:00:00.000-07:00
Location: San Francisco, CA

Saturday Robotics

San Francisco, CA

Past Event

Please click on the button below to join the waitlist. You will be notified if additional spots become available.

About Event

Robotics & World Models Reading Club 30: What Scale Does Not Fix: Representation and Architecture in Robot Manipulation. SF 9/26

A high-signal reading group for AI researchers & builders pushing the frontiers of robotic world models, WAMs, and embodied intelligence. In our previous sessions, we brought together researchers and engineers from Boston Dynamics, Google DeepMind, NVIDIA, Stanford, UC Berkeley, Physical Intelligence, Tesla, Generalist, Rhoda AI, and leading Bay Area robotics startups.

Hosted by Junfan Zhu & Aurora Feng.

Support Saturday Robotics Inc: https://donate.stripe.com/28EcN52rjgeY1fJboYgEg00

Reading Club 30's Core Theme

What Scale Does Not Fix: Representation and Architecture in Robot Manipulation

Keynote: Devol Robots (Elijah Yi Herng Ong - CTO)

Robot foundation models have scaled rapidly over the past two years, yet scale alone has not produced reliable physical interaction. This talk presents two architectures from Devol Robots that attack the problem from opposite ends: what a model represents, and how that model is organized.

The first is the Interaction World Model (IWM 1.0), which addresses representation. Built on Interaction Geometry, a Riemannian framework that keeps robot sensorimotor signals on the manifolds where they physically live rather than flattening them into Euclidean vectors, IWM infers compliance from the shape of a demonstration itself and commands the robot through an impedance controller whose stiffness and damping the model predicts, with no force or torque signals anywhere in training. On contact-rich tasks from socket insertion to sub-millimeter lens seating it reaches 92% average success against 38% and 50% for leading VLA and world model baselines, using roughly an order of magnitude fewer action parameters.

The second is Devol-ONE, which addresses architecture. Existing world-action models keep prediction and policy structurally separate and connect them only through a predicted output. Devol-ONE instead runs vision-language reasoning, latent world prediction, and action generation as three streams of a single autoregressive Mixture of Transformers that exchange information at every layer, so imagination shapes execution continuously rather than at one bottleneck. It reaches 98.4% on LIBERO and 71.4% zero-shot on LIBERO-Plus without training on the perturbed data, and holds its performance on RoboTwin 2.0 under environment randomization while a strong pixel-space baseline collapses from 76.6% to 2.7%.

Elijah will show how these two lines converge: a representation that captures the physics of contact, carried by an architecture that reasons about the future while it acts. Together they define what a deployable and reliable manipulation system looks like in industrial settings, and he will demonstrate what each looks like running on production hardware today.

Keynote 2

Omni-Reason: A General Video Model That Could Solve 100 Computer Vision Tasks

Hokin Deng, Research Scientist @ Harvard, CMU

Emergence of reasoning through scaling laws has hallmarked artificial intelligence. Before language reasoning, there has been 1000 tasks in the field of natural language processing. But now, every task becomes a sub-task of language reasoning. In the same way, we see the arrival of a generalized video reasoning model that could solve many computer vision tasks as an important milestone in the scaling law of vision models. In this belief, we present Omni-Reason. First, Omni-Reason-Data, 9.33 millions samples training data corpus with an end-to-end data infrastructure that scales mid-training data to at least 10,000 samples per task across 224 computer vision tasks. Second, Omni-Reason-Bench, a carefully curated video-to-video benchmark for assessing emergent reasoning in video models. Third, Omni-Reason-1, the first generalized video reasoning model to achieve above-chance performance on all 100 tasks. Omni-Reason-1 is trained on Amazon Trainium, for which we built a training stack from scratch that is both fast and compute-efficient. We further report a non-trivial trade-off between data mixtures. We open-source the entire pipeline, data infrastructure, training data, benchmark, training stack, and training recipes, to accelerate the arrival of general visual reasoning models.

Pre-Readings

www.devolrobots.ai/whitepaper

Location

Date & Time

Saturday, September 26, 2026 | 2:00 PM – 5:00 PM

Join Discord Community

Join Discord Server

Follow Saturday Robotics on X

/saturdayrobotic

Agenda

2:00 PM – 2:30 PM Door Opens & Social

Food 😋, beverages🧋 and UNLIMITED strawberries 🍓 (our official reading club fruits ☺️😄).

2:30 PM – 3:30 PM Keynote by Devol Robots, Hokin Deng

YouTube Recording: TBD (We are looking for recording volunteers)

3:30 PM – 5:00 PM Q&A, open-floor roundtable (10–20 min per topic) on spotlight papers or any paper you’d like to highlight. Feel free to share why the paper matters and its technical details.

Future events

#iros-reading-club-31-0928: 🍾 IROS 2026 x Saturday Robotics — Robotics Research Night | Reading Club 31. Pittsburgh 9/28

Session 31 Luma: https://luma.com/tzbw7n61

Past events

#reading-club-28-0912: Booster T2: The Next Frontier of Open Humanoid Robotics

Session 28 Luma: https://luma.com/00z3oxw6

#reading-club-27-0905: Rethinking Robot Development: Co-Designing Morphology, Sensing, and Learning

Chenyang Ma (Applied Intuition, UNC)

Session 27 Luma: https://luma.com/jm9jceqr

#reading-club-26-0829: Sunnyvale 8/29

Session 26 Luma: https://luma.com/enowqo5v

#reading-club-25-0822: Contact-Rich Robot Learning from Human Videos and Tactile. SF 8/22

Kelin Yu (Maryland, Amazon FAR)

Session 25 Luma: https://luma.com/yl76re1b

#reading-club-24-0820: Saturday Robotics x EAI @ WRC Beijing | PHYSICAL AI UNPLUGGED —— 与 Physical AI 研究者的一场非正式夜谈

• 李晓帆 — Head of World Models, X Square Robot (自变量机器人)

• Tongzhou Mu — Research Scientist, Rhoda AI

• Haoyi Niu — University of California, Berkeley

• Yufei Wang — Member of Technical Staff, Genesis AI

• Shibo Zhao — Research Scientist, Amazon FAR

• 郑思鹏 — Technical Partner, Being Beyond

• Tianle Zhang — Embodied AI Researcher, JD Explore Academy

• Yihang Li — Embodied AI Researcher, JD Explore Academy

Session 24 Luma: https://luma.com/jezexfza

#reading-club-23-0815: Engineering Robotic Simulators for Evaluation and Beyond. SF 8/15

Kaifeng Zhang (Columbia, World Labs)

Session 23 Luma: https://luma.com/non72bev

#reading-club-22-0808: ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow. SF 8/8

Haoyi Niu (UC Berkeley)

Session 22 Luma: https://luma.com/7nnl4a5r
Reading Club 22 Review: https://x.com/junfanzhu98/status/2086482247344079100?s=20
Reading Club 22 Recap: https://x.com/junfanzhu98/status/2086484540013560280?s=20

#reading-club-21-0801: Vision-Language-Kinematics Supervision for Perception-Based Humanoid Loco-Manipulation — SF 8/1

Yen-Jen Wang (UC Berkeley, Amazon FAR)

Session 21 Luma: https://luma.com/4xkibxbh
Reading Club 21 Review: https://x.com/junfanzhu98/status/2083948325112533456?s=20
Reading Club 21 Recap: https://x.com/junfanzhu98/status/2083950393734975839?s=20

#reading-club-20-0725: Agentic Robotics Models (ENPIRE and Cap-X), Mountain View 07/25

Haoru Xue (UC Berkeley)

Kris Hauser (Samsung Research America, Robot Intelligence Lab)

Session 20 Luma: https://luma.com/5ltk12w5
Reading Club 20 Review: https://x.com/junfanzhu98/status/2081420671180284190?s=20
Reading Club 20 Recap: https://x.com/junfanzhu98/status/2081423890988044794?s=20

#SIGGRAPH-reading-club-19-0722: SIGGRAPH x Saturday Robotics — World Models for Robotics: Bridging Graphics, Simulation & Physical Intelligence | Reading Club 19, LA 07/22

Session 19 Luma: https://luma.com/yh2212ac
Reading Club 19 Review: https://x.com/junfanzhu98/status/2081923681627079055?s=20
Reading Club 19 Recap: https://x.com/junfanzhu98/status/2081924824709124174?s=20

#reading-club-18-0718: Causal World Models For Real-World Intelligence. SF 07/18

Guanming Wang & Bill (General Instinct, YC P26)

Feng Fan (UCSD, Aether AI)

Session 18 Luma: https://luma.com/f74eguvr
Reading Club 18 Review: https://x.com/junfanzhu98/status/2078738476602884502?s=20
Reading Club 18 Recap: https://x.com/junfanzhu98/status/2078747463058694625?s=20
Autonomous Panel Moderation: Rebuilding the Factory: Physical AI on the Production Line: https://x.com/junfanzhu98/status/2078385948887490814?s=20

#reading-club-17-0711: Soft Tactile-Centric Multimodal Intelligence Toward Safe and Dexterous Manipulation. SF 07/11

Quan Luu, Purdue.

Session 17 Luma: https://luma.com/e53zawq2
Reading Club 17 Review: https://x.com/junfanzhu98/status/2076205508646850819?s=20
Reading Club 17 Recap: https://x.com/junfanzhu98/status/2076207394426917187?s=20

#private-lunch-icml-0709: Saturday Robotics x ICML Private Lunch (Seoul)

Private Lunch: https://luma.com/khqg11gt

#reading-club-16-0704: The Embodied AI Hardware Stack — Supply Chain, Sensors, and the Data Flywheel — SF 07/04

Jerry Huang, Robotics Center of Silicon Valley.

Session 16 Luma: https://luma.com/cgzyfpeb
Reading Club 16 Review: https://x.com/junfanzhu98/status/2073666442512547957?s=20
Reading Club 16 Recap: https://x.com/junfanzhu98/status/2073667174678589620?s=20

#reading-club-15-0627: Scaling Touch: Flexible Tactile Skin for Dexterous Manipulation

Binghao Huang, Columbia, Amazon FAR.

Session 15 Luma: https://luma.com/7lts5ppf
Reading Club 15 Review: https://x.com/junfanzhu98/status/2071129728984273105?s=20
Reading Club 15 Recap: https://x.com/junfanzhu98/status/2071132114226139489?s=20

#deep-tech-week-14-0625: Deep Tech Week Research Night

SPEAR: A Simulator for Photorealistic Embodied AI Research. (ECCV 2026 accepted) by Mike Roberts, Senior Research Scientist, Adobe Research.

Bogdan Cristei, Venture Partner at SHACK15 Ventures.

Simone Totaro, CTO at Saturn Dynamics.

Shumo Chu, CEO at General Intelligence Labs.

Margaret Zhang, CEO at ThirdBrain Labs.

Session 14 Luma: https://luma.com/l1g9c2l1
Deep Tech Week 14 Review: https://x.com/junfanzhu98/status/2070417784560066592?s=20
Deep Tech Week 14 Recap: https://x.com/junfanzhu98/status/2070420270964514923?s=20

#reading-club-13-0620: HumanEgo: Train Robot Policy from 30 min Egocentric Videos — SF 0620

Session 13 Luma: https://luma.com/6vkhxnum
Reading Club 13 Review 1: HumanEgo: Zero-Shot Robot Learning, Human Egocentric Video, Leo Wang, Amazon FAR & UMaryland https://x.com/junfanzhu98/status/2068603103138713824?s=20
Reading Club 13 Recap 1: https://x.com/junfanzhu98/status/2068605511549743127?s=20
Reading Club 13 Review 2: Causal World Models: Biwei Huang, Aether AI & UCSD
https://x.com/junfanzhu98/status/2068743219685560453?s=20
Reading Club 13 Recap 2: https://x.com/junfanzhu98/status/2068746229643936177?s=20

#reading-club-12-0613: Origami Robotics (YC W26) on Dexterity

Session 12 Luma: https://luma.com/5w7c1t2a
Reading Club 12 Review: https://x.com/junfanzhu98/status/2066275974988337178?s=20
Reading Club 12 Recap: https://x.com/junfanzhu98/status/2066278639554245097?s=20

#cvpr-denver-11-0606: 🤖🥘 Saturday Robotics x Manycore Tech x Neural Motion | CVPR 2026 Denver Research Night | Robotics & World Models Reading Club 11

Junfan Zhu & Aurora Feng, Founders of Saturday Robotics

Anthony Zhao, Head of North America at Manycore Tech SpacialVerse

Aurora Feng, Founder at Neural Motion. NM-GenET.

Max Zhaoshuo Li, Robotics and World Model Tech Lead at NVIDIA Cosmos. Cosmos 3.

Xiaofan Li, World Model Tech Lead at X Square Robot. WALL-WM.

Zesen Zhao, University of Michigan. Test-Time Scaling for World Action Models via Zero-Shot Geometric Verification.

Pengyi Liao. VGGT-Ω: From 3D Reconstruction to Scalable Spatial Representation.

Jie Wang, University of Pennsylvania, GRASP Lab. Toward a Robotics MMLU: Lessons from Sim & Real Evaluations of Generalist Policies.

Gordon Qian, Senior AI Researcher at Snap. Diffusion-DRF: Free, Rich, and Differentiable Reward for Video Diffusion Fine-Tuning.

YouTube livestream: Watch on YouTube
CVPR Denver Research Night Luma: https://luma.com/zamm9g2g
CVPR Denver Research Night Lightning Talks Review: https://x.com/junfanzhu98/status/2065102150418788581?s=20
CVPR Denver Research Night Lightning Talks Recap: https://x.com/junfanzhu98/status/2065104616497524843?s=20
CVPR Hot Takes Review: https://x.com/junfanzhu98/status/2065234892653547904?s=20
CVPR Hot Takes Recap: https://x.com/junfanzhu98/status/2065236892988416166?s=20

#reading-club-10-0530: Bringing Robots to Life — Learning Humanoid Instincts from the Body Up | San Francisco 0530

Haochen Shi (Stanford, co-advised by Karen Liu & Shuran Song)

Session 10 Luma: https://luma.com/czz76qe1
Reading Club 10 Recap: https://x.com/junfanzhu98/status/2061145697693683878?s=20

#private-dinner-01-0529: Robo Plov x Saturday Robotics

Private Dinner 01 Luma: https://luma.com/3rzqwond
Private Dinner 01 Review: https://www.linkedin.com/posts/junfan-zhu_saturday-robotics-was-excited-to-host-its-activity-7466384182190182400-Zgnv?utm_source=share&utm_medium=member_desktop&rcm=ACoAABxP-p0BpUNGDf347aKh_1uJAPzG4er0As8

#reading-club-09-0523: CVPR Warm-up & Founders Spotlight — DeltaWorld + VisuoTactile Dexterous Hands

Tommie Kerssies (Amazon Frontier AI & Robotics)

Arjun Subramaniam (Factory Intelligence)

Session 09 Luma: https://luma.com/wooiz0bf
Reading Club 09 Review: Part 1: Robotics & World Model Reading Club 9.1: CVPR Warm-up— A Frame is Worth 1 Token: DeltaToken. https://x.com/junfanzhu98/status/2058449627184267621?s=20
Reading Club 09 Review: Part 2: Robotics & World Model Reading Club 9.2: Tactile Sensor & Reliable Manipulation in Production. https://x.com/junfanzhu98/status/2058461947637694948?s=20

#reading-club-08-0516: Embodied Human Data as the “Internet of Motion and Behavior”

Ryan Punamiya (NVIDIA Gear, Georgia Tech)