Mifeng studies real-robot data quality across 207,000 hours
11/20/2026 — 11/20, 19:20·1 sources (all)·1 reports (all)
Key takeaways
- Shanghai-based embodied AI firm Mifeng has published a study on real-robot data quality built on 207,000 hours of collection logs and more t…
Background & analysis
On November 20, 2026, Shanghai MiFeng Embodied Intelligence Technology Co., Ltd. officially released a study on the quality of real-robot data. The research draws on 207,000 hours of real-robot collection records and more than 20,000 task records, and validates findings on actual robots using a fine-tuned π0.5 model. π0.5 is a general-purpose robotic foundation model developed by Physical Intelligence; the MiFeng team fine-tuned it on data the team collected through its own teleoperation sessions before running the real-robot evaluations.
Methodologically, the team brings together three previously separate research threads — robot body adaptation, teleoperation systems, and imitation-learning data quality — and discusses them within a single real-robot data production pipeline, quantifying part of their impact through controlled experiments. The core finding is that a robot completing a task successfully does not guarantee that the resulting data will be useful for training, and even successful demonstrations can teach the model the wrong behavior.
The related paper was submitted to arXiv on November 17, 2026, and is authored by MiFeng's real-robot manipulation team. It remains a technical report that has not yet undergone peer review. The contribution statement divides participants into categories including core contributors, team leads, contributors, and Sponsors. The sponsor list includes Yao Maoqing, Chairman and CEO of MiFeng, Co-founder, Co-President and President of the embodied business system of Zhiyuan (AgiBot), along with Ren Guanghui, Vice President of Zhiyuan's Embodied Research Center, though the nature of their support is not specified. MiFeng is a data-services company incubated by Zhiyuan Robot, and the paper's title page also lists a Zhiyuan robot teleoperation website domain. As of now, the work is available only as an unreviewed technical report.
AI-generated from 1 reports · updated 2 days ago
- Heat index
- 32
- All sources
- 1
- All reports
- 1
- First seen
- 2 days ago
Reports on this story headlines open the original
Shanghai embodied AI company Mifeng released a real-robot data quality study based on 207,000 hours of real-robot collection records and over 20,000 task records, validated on physical robots with a fine-tuned π0.5 model. The study concludes that a robot completing a task successfully does not guarantee good training data, and successful demonstrations can also teach the model bad behavior. The paper was submitted to arXiv on November 17 as a non-peer-reviewed technical report.
华尔街见闻AI score 57
Other stories people are talking about
- Heat index 1143RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1011Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
- Heat index 622Anthropic IPO total raise reaches $93.5 billion after greenshoe exercise20 sources
- Heat index 596RisingThinking Machines Lab Open-Sources Inkling 2 Mini with 32GB Model Card Recommendation10 sources
- Heat index 490RisingSEC probes Nvidia accounting practices as shares post seven-week slide9 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

