Zhejiang University
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 5
- Embodied AI challenge and 2026 China Robot Competition conclude in Hangzhou
The first Global Embodied Intelligence Application Innovation Challenge and the 2026 China Robot Competition closed in Hangzhou on the afternoon of October 18. According to an article QbitAI republished from the National Embodied Intelligence Application Pilot Base, the three-day event drew more than 2,000 teams and nearly 5,000 contestants from universities including Tsinghua, Zhejiang and Shanghai Jiao Tong, competing through real-robot operation and live-scenario tasks. The competition awarded 59 championships, 57 runner-up places, 52 third places and 12 enterprise-track team awards, with tasks spanning flexible sorting, air-ground coordination, home service, medical assistance, unknown-terrain exploration and dexterous manipulation. Executives from 10 listed companies including Sunyard, Wasu Media and Unitree Robotics presented awards, and a hiring event with roughly 50 firms such as Unitree, Alibaba DAMO Academy and Hikvision offered more than 500 internships; champion teams received an "internship gold card."
量子位（原生 RSS） · 🔥 0 - Single luminescence image plus lightweight AI maps PV module power loss at about 0.5% error
A Zhejiang University team led by Dongchen Lan published research in Matter & Light, selected as the journal's October cover: a lightweight AI model first judges from a single luminescence image whether a module's degradation is dominated by defect recombination or by series-resistance losses, and a device physics model then reconstructs local electrical behavior and the power-loss distribution. Validated on about 300 field-recovered crystalline-silicon modules from different production batches and operating environments, the reconstructed maximum power differed from laboratory standard current-voltage measurements by roughly 0.5% overall. The team also tested the method at a 186 kW plant with 342 modules, capturing luminescence images at night from a drone-mounted camera; in one representative area, 24 modules averaged 7.53% power loss, with some severely degraded cell regions near 25%. The team says next steps focus on using the method for routine plant operations and maintenance inspection, and the paper states the approach can extend to perovskite and tandem modules but requires redesign and revalidation.
MIT科技评论中文 · 🔥 0 - Zhejiang University proposes EMem-Bench for embodied agent memory
Zhejiang University's OmniAI team introduced EmbodiedMemory-Bench (EMem-Bench), a benchmark with four task types—passive observation, dynamic tracking, interaction failure and experience generalization—and 2,554 executable samples testing whether embodied agents can use past observations, state changes and action feedback in later tasks. The team also proposed an external memory system, Embodied-Memorizer (EMem), and fine-tuned Qwen3-VL-8B into EMem-8B; in the authors' evaluation, EMem raised average success from 35.3% to 58.9% with a GPT-5.4-mini backbone. Under the paper's full-context setting, Gemini-3-Flash led at 64.2% average success and GPT-5.4 reached 56.8%. The report says the benchmark drew nearly 30,000 Hugging Face downloads in about three weeks, with results mainly from simulation.
新智元 · 🔥 0 - China's 'face-selling' AI video gigs raise privacy risks, experts warn
Face-authorization platforms in China have been advertising offers such as a bank employee earning nearly 2,500 yuan in three days by "selling their face," according to a Science and Technology Daily report republished by cnBeta: users upload facial photos and sign licensing agreements letting platforms use their likeness in AI short dramas, commercials and similar content, with the platform taking a cut. Wu Shenkuo, a professor at Beijing Normal University, said such deals involve both portrait licensing and processing of sensitive facial information, and that many platforms' agreements are crude and carry legal and data-security risks. Li Wenlong of Zhejiang University's Institute for Digital Rule of Law said text-to-video is changing content production and face licensing could become a niche market, but the main risk is that one authorization may extend to unforeseen uses such as sublicensing, long durations or model training, while AI's ability to split and merge facial features makes infringement hard to detect and pursue. The report did not verify whether the income claims are real.
cnBeta · 🔥 0 - AgiBot sends 1,000+ robots to 20-plus universities in ¥200M program
IT Home reported on October 16 that AgiBot (Zhiyuan) announced partnerships with a first batch of more than 20 universities, including Peking, Tsinghua, Zhejiang, Shanghai Jiao Tong, Xi'an Jiaotong and South China University of Technology, to build an industry-education platform for embodied AI. AgiBot said it will supply over 1,000 of its own robots — Genie, Lingxi and Expedition humanoids plus Kuotuo quadruped robots — and give schools access to its AIMA development platform (LinkSoul, LinkCraft and Genie Studio) plus paid compute, with robot hardware and software memberships valued at about ¥200 million. A Xinzhiyuan report on October 19 added that Tsinghua's electronics engineering department has set up a teaching lab with five Genie G1 robots and VR teleoperation gear, and now runs a 2-credit, 45-hour course in which students must deploy their own trained models on real robots.
IT之家 AI · 🔥 0
Experience and discussion from the community
Share my Zhejiang University experienceAsk about Zhejiang University
Nobody has shared their experience with Zhejiang University yet.