As the prelude and extension of 2026 AGIBOT INNOVATION DAY, starting October 22, AGIBOT will launch "AI WAVE," consecutively releasing seven achievements including datasets, an operation platform, robot bodies, an Agent framework, a simulation platform, a foundation large model, and a collective experience learning paradigm, centrally presenting a full-stack embodied AI technology layout.
Release is not the finish line—all released technologies will ultimately undergo validation in real-world scenarios. From November 2 to 3, 6 G2 robots will enter convenience stores and, through 24-hour uninterrupted livestreaming, go through four complete shifts—lunch peak, dinner peak, late-night duty, and breakfast preparation—to complete a "probation review."
In the past few months, the debate over "as general large models become increasingly powerful, does embodied intelligence still have value" has swept the industry.
Our answer has never relied on debate: the large-scale deployment of robots depends on full-stack coordination, not single-point breakthroughs. Completing a single task requires coordination between the model and the body; long-term stable operation requires environmental adaptation and exception handling; scaled application requires low training cost, fast deployment, and capability reuse. In the physical world, data, tools, bodies, systems, and models—not a single link can be missing for anything to run.
That is why we have this AI WAVE: we put every link on the table, and then place the robots in a real convenience store to "go on duty" for 24 hours.
We hope to use full-stack embodied AI technology as a bond to connect the industrial ecosystem, accelerate the deployment and evolution of physical AI, and drive robots to truly enter the real world, serve real scenarios, and create real value.
Seven days, seven achievements, one "why" per day
10/22 | AGIBOT WORLD 2026 Phase 4 "Physical Reasoning" dataset
The first batch of 150 hours of real-machine data focuses on decision-making—driving robots from "learning to operate" toward "autonomously completing tasks."
10/23 | Genie Studio 2.0
A one-stop full-process embodied intelligence operation platform—enabling robots to move from "usable at a single point" toward "continuous evolution, standardized delivery, and scale replication."
10/26 | G2 Max
A safety-standard heavy-load embodied industrial robot—achieving both heavy load capacity and sub-millimeter precision operation.
10/27 | Genie Master
Connecting Agents with the physical world—realizing Physical RSI (embodied recursive self-improvement) for the first time in real business scenarios.
10/28 | Major upgrade of Genie Sim's self-developed physics engine
Comprehensively overcoming the bottleneck of flexible simulation—real-machine-to-simulation consistency exceeding 95% for complex tasks.
10/29 | Foundation large model
Redefining the evolutionary boundary of embodied intelligence—stay tuned.
10/30 | Collective Experience Learning (CEL)
The first robot learning paradigm to complete prototype validation on a hundred-level real-machine cluster—enabling robots to move from "relying on human demonstration" toward "collective autonomous growth."
Final stop: convenience store, 24 hours, real work
From November 2 to 3, 6 G2 robots will enter a real convenience store: stocking shelves, replenishing goods, organizing merchandise, preparing hot food, collecting trash, making coffee, and greeting customers—from 10 a.m. to 10 a.m. the next day, four shifts running continuously, livestreamed throughout.
Starting October 22, witness technology entering real scenarios.
What we are building is a closed loop that connects learning, action, and continuous evolution: accumulating data from real interaction, precipitating experience through continuous operation, and then feeding new experience back into the next round of learning.
Every step of bringing robots into the real world is supported by a full-stack technology system.
