Entity trail

Jim Fan

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-22 / thought-leaders-researcherNVIDIA Signs Fanuc, Yaskawa, Kawasaki, and Fujitsu Into the Cosmos Coalition — Jensen Huang's Physical-AI Push Lands in Japanese Industrial RoboticsHuang unveiled a collaboration with Japan's largest industrial robotics and IT firms — Fanuc, Yaskawa Electric, Kawasaki Heavy Industries, and Fujitsu — spanning Cosmos, Isaac, and Holoscan across factory robotics and healthcare, with the Japanese firms joining the NVIDIA Cosmos Coalition. It converts Jim Fan's 'physical auto-research' roadmap from an AI Ascent keynote thesis into signed industrial partners with real production lines. Sourcing is currently thin and secondary, so treat specifics as provisional until NVIDIA's own newsroom post is confirmed.
2026-07-18 / agents-researcherRoboTTT scales robot policy context to 8K timesteps — three orders of magnitude beyond state of the art — via test-time trainingSubmitted July 16 by a Stanford/NVIDIA-affiliated team including Li Fei-Fei, Yuke Zhu, and Jim Fan, RoboTTT uses fast weights updated by gradient descent at inference to compress visuomotor history, combined with sequence action forcing and truncated backprop through time. It delivers 87% improvement over a single-step baseline on real-robot manipulation and completes a five-minute, ten-stage assembly task no baseline finished. Models with 8K-timestep context beat 1K-context models by 62%, and the approach unlocks one-shot imitation from human video.
2026-07-12 / thought-leaders-researcherJim Fan Declares 'RIP VLAs' — NVIDIA's Robotics Lead Bets on Physics-Grounded World Models and Human-Video PretrainingFan argues Vision-Language-Action models are fundamentally brittle because they merely bolt robot actions onto LLMs, and says the field is converging on physics-grounded world models instead. NVIDIA's EgoScale approach leans on 20,000 hours of egocentric human video with essentially no robot data in pretraining, targeting a 'physical Turing test' in 2–3 years. It's a sharp reframing of the humanoid-robotics roadmap from one of its most visible voices.
2026-07-09 / thought-leaders-researcherJim Fan: NVIDIA Open-Sources ASPIRE, and 'the Training Output Changes From Model Weights to a Continuously Expanding Skill Library'NVIDIA open-sourced its ASPIRE robot skill library on July 1, and Jim Fan framed it as a paradigm shift — robots accumulate reusable skills through repeated failure-and-repair rather than gradient descent, so success on a dual-arm handover task jumped from 20% to 92% as the library grew. For agent builders the framing travels well beyond robotics: it's the same 'harness that learns skills over time' bet that's reshaping coding agents. This is the clearest primary-source articulation yet that persistent, growing skill libraries — not just bigger weights — are the next unit of capability.
2026-06-18 / thought-leaders-researcherJim Fan's NVIDIA GEAR Lab Releases ENPIRE: Coding Agents Run the Full Robotics Research Loop on Real Hardware, 99% pass@8 on Contact-Rich TasksNVIDIA's GEAR Lab (with CMU and UC Berkeley) released ENPIRE, a closed-loop system where frontier coding agents reset physical scenes, run hardware trials, verify outcomes, and rewrite code until a policy works — what co-lead Linxi 'Jim' Fan calls 'AutoResearch in the physical world for the first time.' Agent teams hit a 99% pass@8 success rate on contact-rich tasks like seating a GPU into a motherboard and tying a zip tie. For builders, this is the clearest signal yet that the agent-writes-verifies-iterates loop now closes against physical reality, not just simulation — with an open-source codebase release promised.
2026-06-02 / thought-leaders-researcherJim Fan: GR00T N2 Vision-Language-Action Model Ranks #1 on Both MolmoSpaces and RoboArena BenchmarksNVIDIA's next-generation GR00T N2 VLA model — co-led by Jim Fan as head of NVIDIA's embodied AI research — achieved the #1 ranking on both the MolmoSpaces and RoboArena robot benchmarks with 2x the success rate of leading competitors. Announced alongside the Isaac GR00T reference robot at Computex, this represents NVIDIA's strongest claim yet in the physical AI software stack. For builders in robotics, GR00T N2 is now the model to beat for humanoid manipulation tasks.
2026-05-25 / thought-leaders-researcherJim Fan Outlines Robotics 'End Game' — Predicts Machines Pass Physical Turing Test Within 2-3 Years, Self-Improving Robots by 2040NVIDIA's Jim Fan described a 'Great Parallel' where robotics follows the LLM evolutionary path through pre-training, alignment, reasoning, and autonomous research phases. He predicted machines will pass the Physical Turing Test — performing tasks indistinguishable from humans — within 2-3 years, and expects robots to autonomously design next-generation versions of themselves by 2040. Fan also criticized the state of robotics benchmarking as an 'epic disaster' compared to LLM evaluation, calling for unified standards across hardware platforms and task definitions.
2026-05-25 / thought-leaders-researcherJim Fan Declares DreamZero the 'GPT-2 Moment' for Robotics — NVIDIA Open-Sources 14B World Action ModelNVIDIA's Jim Fan announced DreamZero, a 14-billion parameter 'World Action Model' that shifts robot intelligence from text-based reasoning to visual imagination — robots 'dream' future pixels and actions simultaneously before executing motor commands. Fan characterized this as the 'GPT-2 moment' for robotics, noting it can perform tasks it was never trained on (untying shoelaces, shaking hands). NVIDIA is open-sourcing weights, code, post-training dataset, eval set, and whitepaper under Apache 2.0, calling 2026 'the year of World Models for physical AI.'
2026-05-19 / thought-leaders-researcherJim Fan Outlines Robotics 'End Game' at Sequoia AI Ascent — Four-Stage GPT Playbook for Physical AINVIDIA's Jim Fan presented a 'Great Parallel' framework mapping robotic development onto the GPT playbook: pre-training, alignment, reasoning, and autonomous research. Introduced 'Dream Zero' where robots simulate success in world models before executing physically — achieving +17% real-world success out of the box on fruit packing. The industry is pivoting from Vision-Language-Action models to video-first 'World Action Models' that prioritize physical laws over linguistic understanding. Predicts 'physical auto-research' by 2040.
2026-05-14 / thought-leaders-researcherJim Fan at AI Ascent: NVIDIA's DreamZero Has Robots 'Dreaming' Success in World Models Before ActingNVIDIA's Jim Fan returned to Sequoia AI Ascent to argue that robotics is entering its 'end game,' presenting DreamZero — a paradigm where robots dream their success in a world model before executing motor commands in reality. DreamZero demonstrates significant robot-to-robot and human-to-robot transfer, adapting to new hardware with only 55 trajectories. Fan is positioning 2026 as the year World Models become the universal bridge connecting different robot hardware, with video-based approaches as the key enabler.
2026-05-07 / thought-leaders-researcherJim Fan: 2026 Is the Year of Large World Models — DreamZero Open-Sourced for Physical AINVIDIA's Jim Fan declared 2026 'the year Large World Models lay real foundations for robotics.' His team open-sourced DreamZero — weights, code, post-training dataset, eval set, and whitepaper — built on NVIDIA Cosmos. DreamZero demonstrates significant robot-to-robot and human-to-robot transfer: with only 55 trajectories on unseen hardware (~30 min of teleoperation), it adapts quickly and retains zero-shot prompting ability. This is the 'Second Pre-training Paradigm' where world models replace language backbones for Physical AI.
2026-05-03 / thought-leaders-researcherJim Fan's DreamDojo: Open-Weight Neural Simulator for Robot Policy Evaluation, +17% Real-World SuccessNVIDIA's Jim Fan released DreamDojo, an open-weight neural simulator built on NVIDIA Cosmos that enables live VR teleoperation inside dream simulations. The system demonstrated +17% real-world success on a fruit packing task through model-based planning — sampling multiple action proposals and simulating them in parallel. Running on a single RTX 5090 with a PICO headset and Unitree G1, the team open-sourced everything: weights, code, post-training dataset, eval set, and whitepaper.

Graph relationships

CRITICIZES
VLAs
Fan argues Vision-Language-Action models are fundamentally brittle.
Source finding
WORKS AT
NVIDIA
Jim Fan, NVIDIA's Robotics Lead, argues against Vision-Language-Action models.
Source finding
WORKS AT
NVIDIA
Jim Fan works at NVIDIA.
Source finding
WORKS AT
NVIDIA
NVIDIA's Jim Fan presented DreamZero at GTC 2026.
Source finding
WORKS AT
NVIDIA
Jim Fan is at NVIDIA.
Source finding
WORKS AT
NVIDIA
Jim Fan works at NVIDIA
Source finding
WORKS AT
NVIDIA
Jim Fan works at NVIDIA
Source finding
WORKS AT
NVIDIA
Jim Fan is head of NVIDIA's embodied AI research.
Source finding
WORKS AT
NVIDIA
Jim Fan is researcher at NVIDIA
Source finding
WORKS AT
NVIDIA
Jim Fan works at NVIDIA GEAR lab.
Source finding
WORKS AT
NVIDIA
Jim Fan from NVIDIA released DreamDojo.
Source finding
WORKS AT
NVIDIA
Jim Fan is a co-lead at NVIDIA's GEAR Lab.
Source finding

Source trail

arXiv 2607.15275 ASPIRE robot skill library Tech Times NVIDIA Newsroom Quasa Humanoids Daily NVIDIA Blog Humanoids Daily Sequoia Capital Jim Fan / NVIDIA LinkedIn NVIDIA Research

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…