AgiBot to launch ‘AI WAVE’ on Oct. 22; six G2 robots to operate around the clock at convenience store starting Nov. 2

AgiBot announced Oct. 21 that it will launch ‘AI WAVE’ on Oct. 22. The seven planned releases are datasets, a task platform, robots, an agent framework, a simulation platform, a foundation model and a collective experience-learning paradigm. The company also said six G2 robots are scheduled to take part in a 24-hour livestream at a convenience store from Nov. 2 to 3 and undergo field testing.

AgiBot’s plan brings together datasets, a task platform, simulation and a foundation model rather than focusing on robots alone. If the planned 24-hour demonstration at a convenience store goes ahead, it will put the development and evaluation of robots designed for continuous operation in real-world settings to the test.

AgiBot’s announcement stands out not as a single product launch but as a coordinated lineup of data, a task platform, robots and a learning framework. In particular, its seven planned releases include a dataset using 150 hours of data from physical robots, a simulation platform billed as matching real robots with virtual environments, and collective experience learning, all slated for the same week. This suggests the company is emphasizing a cycle of data collection, task execution and retraining, not just model performance. The 24-hour convenience-store demonstration is intended to test whether that cycle can operate in a real-world environment. Our related story, AgiBot invests in Xinji Infinity, showed that the company is expanding its involvement beyond robots to surrounding technologies and the broader ecosystem. But the focus this time is not on investment; it is on how the planned technologies will be connected and put into real-world operation. The investments and partnerships, as well as the successive releases of datasets, platforms and robots, are part of the company’s broader effort to build up the supply side of physical AI in stages. From an industry perspective, the key test is the real-world condition of “24 hours at a convenience store.” Handling stocking, replenishment, merchandise organization, hot-food preparation, cleaning, coffee service and greeting customers across four shifts requires more than a one-off demonstration: perception, task planning, anomaly response and continuous operation must all work together. AgiBot’s simultaneous release of datasets, a task platform, simulation and a foundation model appears aimed at demonstrating an integrated operating system rather than isolated functions. The announcement, however, does not yet make clear how autonomously the six G2 robots will operate, which tasks will require human support or what scope the 150 hours of physical-robot data will cover. Production volumes, uptime in sustained operation and intervention procedures in the event of failures will be important points to watch.

For Japanese companies, an integrated approach to collecting data from physical robots, running a task platform and conducting virtual validation could become a competitive factor. In particular, when targeting real-world use in retail and service settings, the presence of an operating model that includes continuous learning may set companies apart, beyond components and systems integration alone.