December 18, 2025

HanabiAI Co., Ltd.

HanabiAI Co., Ltd. (Headquarters: Tokyo; Representative Director: Kagaya Sano; hereinafter "the Company") visited Robot Innovation Week 2025, held at Akihabara UDX from December 16 to December 18, 2025, where we conducted research into the latest trends in robot development and engaged in technical exchanges with industry stakeholders.

Robot Innovation Week 2025 is a technical networking event where developers and business operators share knowledge through multiple conferences, special courses, and exhibitions under the theme of “New robot development strategies in the Physical AI era.” The event is organized by TechShare Inc., and its program includes the Unitree Developer Conference, DOBOT User Conference, Learning Robot Conference, AMR UGV Conference, and more.

Purpose of Attendance

The Company believes that the factors determining successful social deployment of robots are not limited to improving standalone performance, but also include development process reproducibility, data acquisition and evaluation, operational design and maintainability, and system integration including peripheral equipment. Through this visit, we aimed to confirm how learning-based development methods and practical considerations for field deployment are being organized and discussed, and to incorporate these insights into the development of our technical foundations.

Key Takeaways from the Company’s Perspective

Based on information obtained at the venue, the Company organized key findings from the following perspectives.

1. The entry point of learning-based development is determined by data acquisition design

Resolution is increasing in data collection methods, including teleoperation-based acquisition and imitation learning data gathering. It became clear that acquisition design and quality management define development speed. Special courses also addressed teleoperation and data collection, underscoring their practical importance.

2. Evaluation is shifting from accuracy to reproducibility and regression detection

As learning and control become more advanced, evaluation design that enables reproducibility under the same conditions and early detection of regressions during updates becomes indispensable. The program structure of the Learning Robot Conference also suggests that the positioning of evaluation and verification is strengthening.

3. The focus of integration is expanding from the robot body to the surrounding ecosystem

Targets for integration are wide-ranging, including robot arms, hands, vision systems, conveyance mechanisms, monitoring, and remote operations. Deployment success increasingly depends on connectivity and operational design. The AMR UGV Conference, which includes conveyance, remote monitoring, and mobile manipulation, is emblematic of this trend.

4. Low-cost automation and lab domains have become established core themes for real-world deployment

At the DOBOT User Conference, low-cost automation and lab automation were highlighted, confirming that designs that lower deployment barriers on the field side continue to be strongly emphasized.

5. Maturation of developer platforms is accelerating deployment speed

As demonstrated by the Unitree Developer Conference program, platform providers have begun presenting not only hardware and tools but also development procedures and learning methodologies. This indicates that improving the developer experience has entered a stage where it can accelerate broader adoption.

Our Future Initiatives

Under our “Robot Brain” vision for embodied intelligent robots, we are progressively organizing and examining the foundational requirements needed to make real-world implementation and operations viable.

First, we are treating the end-to-end process from data acquisition through evaluation as a single, continuous flow. We are clarifying how each stage should connect, including collection, curation, replay, evaluation, and regression detection, and we are studying design approaches that enable a smoother iteration cycle across development and operations.

In addition, we are advancing our review of the compute environment required to make simulation, training, and validation repeatable. In particular, we are focusing on the requirements and system configurations needed to safely deploy deliverables to real robots, with the aim of improving efficiency across the entire verification process in robot development.

We also position observability and maintainability for real-world operations as critical elements. From the early design phase, we are examining how to incorporate considerations such as logging, monitoring, remote updates, and incident response.

Going forward, through continued dialogue and information exchange with partners, we will organize field challenges across multiple robot form factors, including robotic arms, mobile transport robots, bipedal robots, and quadrupedal robots, and will carefully evaluate future validation themes and opportunities for collaboration.

Company Overview

Company Name:HanabiAI Co., Ltd.

Location:Tokyo, Japan

Business Areas:Centered on the development of robot AI brain technologies, we are building the development infrastructure and ecosystem for embodied AI robots

Website:https://www.hanabiai.jp