Current Path：RobotHOT Briefs Tsinghua, Infinigence AI and AGIBOT test collective learning on 96-robot fleet

Tsinghua, Infinigence AI and AGIBOT test collective learning on 96-robot fleet

2026-11-04 Category: Briefs Comments(0)

A team from Tsinghua University, Infinigence AI and AGIBOT has proposed a new paradigm called Collective Experience Learning (CEL) and validated it at scale on a fleet of 96 real robots, the team said. CEL allows robots to learn not only from human demonstrations but also from the successful experience of other robots in the group. When a robot encounters difficulties during a task, it can draw on peers’ past experience to find solutions and improve its performance, according to the team. By sharing and accumulating experience, the robot fleet can improve its capabilities collectively through continued practice, which the team says could support autonomous robot evolution and scaled deployment.

Like(0)
Share To

Related Articles

Comments Get first!

Cancel

Briefs

  • China's RoboScience raises several hundred million yuan in Series A2 round

    RoboScience, a Beijing-based embodied AI startup, said it has completed a Series A2 funding round worth several hundred million yuan. The company will use the proceeds to deepen work in embodied AI, support continued iteration and upgrades of its large models, advance proof-of-concept validation and scaled deployment in logistics, retail and industrial scenarios, and expand its senior talent pool. Founded in 2024, RoboScience focuses on general-purpose embodied AI models and robot bodies.

  • Tsinghua, Infinigence AI and AGIBOT test collective learning on 96-robot fleet

    A team from Tsinghua University, Infinigence AI and AGIBOT has proposed a new paradigm called Collective Experience Learning (CEL) and validated it at scale on a fleet of 96 real robots, the team said. CEL allows robots to learn not only from human demonstrations but also from the successful experience of other robots in the group. When a robot encounters difficulties during a task, it can draw on peers' past experience to find solutions and improve its performance, according to the team. By sharing and accumulating experience, the robot fleet can improve its capabilities collectively through continued practice, which the team says could support autonomous robot evolution and scaled deployment.

  • Unitree A2 wins inaugural Mount Tai robot mountain-climbing race

    The inaugural China Media Group (CMG) World Robot Mount Tai Mountain-Climbing Competition kicked off in Tai'an, Shandong province. Beijing Lingyi Technology, a Unitree Robotics subsidiary, won the legged-robot speed race with its Unitree A2 quadruped robot. The machine completed the course in 41:41, but a two-minute penalty gave it an official time of 43:41. Operating fully autonomously, the A2 relied on real-time perception and decision-making to navigate the mountain's uneven stone steps steadily and swiftly.

  • China computing suppliers dismiss fears of big tech AI spending cuts

    Rumours of weaker AI demand among major tech firms have stoked market worries over AI spending growth, but players in the computing supply chain say they see no signs of pullback and hold ample order backlogs. Xiechuang Data reported no demand reduction signals and expects top domestic internet firms to maintain high capital expenditure next year, keeping computing infrastructure investment buoyant. Xingyun Technology said its disclosed year-to-date computing and storage orders total roughly 20.641 billion yuan.

  • AGIBOT robots complete 1,730 tasks in 24-hour convenience store challenge

    Chinese robotics company AGIBOT said six Genie G2 robots completed 1,730 tasks in a 24-hour livestreamed convenience-store challenge on Nov. 3, drawing over 2 million online views. Tasks included serving customers, delivering goods, making coffee, providing breakfast, restocking, inspections and cleaning. AGIBOT said the real-world test explored embodied AI’s retail potential and built experience for scaled humanoid robot deployment.

  • RoboScience raises several hundred million yuan in A2 funding round

    Chinese embodied-AI startup RoboScience had completed an A2 funding round worth several hundred million yuan. The round was co-led by Amber Capital, Hongruida Investment and Shanghai Semiconductor Industry Investment, with Shuang Capital, 360 Strategic Investment, Guolian Tongbao, Jinding Capital and platforms of several listed companies participating. Proceeds will be used to deepen its presence in embodied intelligence, focusing on continued iteration and upgrading of large models, advancing proof-of-concept validation and scaled deployment in core scenarios including logistics, retail and industry, and further strengthening its high-end talent pool.

  • AGIBOT launches 24-hour convenience-store livestream to test humanoid robots at work

    Chinese robotics company AGIBOT had launched a 24-hour convenience-store challenge livestream titled “Robot Workers Can’t Stop,” with six Genie G2 robots serving as store employees. The robots are handling tasks including greeting customers, making coffee, serving prepared food, picking and restocking goods and cleaning tables, working continuously for 24 hours under real-time online monitoring. Using core technologies including the GO-3 foundation model and the Genie Master embodied agent, AGIBOT aims to validate robots' continuous operation and autonomous service capabilities in a real operating scenario and move humanoid robots from demonstrations to real jobs.

  • DeepSeek Harness team outlines “Super-agent” design for more autonomous AI systems

    The DeepSeek Harness team said in a technical article on its official WeChat account that it has outlined a “Super-agent” design aimed at building more autonomous agent systems. DeepSeek said gains in model capabilities have spurred the emergence of personal agents, but existing mainstream agent designs struggle to support multi-tasking and team collaboration at scale, with multi-task limits tied to human attention. The team said its approach uses a two-tier architecture, event-driven operation, asynchronous multi-channel interaction, separation of storage and computation and an external watchdog to overcome these limits. The system takes over cumbersome session management and context synchronization at the underlying layer, allowing humans to move from execution details to higher-level decision-making and acceptance. DeepSeek said internal Super-agent instances built on the approach are already running continuously, participating in issue reproduction and fixes, code review responses and iteration of DeepSeek Harness itself. When team members report software problems in group chats, the agent can identify potential defects even without being mentioned and try to reproduce the issue and find fixes in the background.
MORE

Hot Topics

RobotHOT - China & Global Robotics Insights

RobotHOT brings China & Global Robotics Insights. Explore robotics, embodied‑intelligence updates, startup news and industry analysis for professionals worldwide.

About USContact US