AIRoA Japan: Building the Foundation for Next-Generation Robotics and AI Technologies with T. Ogata

In this talk

1. 1:15

   Humanoids as a way to study intelligence

   Ogata traces Japan’s humanoid work back about 50 years to an early Waseda robot that stood over two meters tall and took more than 40 seconds to step. His research combines neural networks with developmental robotics: building humanlike systems to better understand cognition.

2. 3:01

   Learning robot skills from people

   Ogata describes early work using human demonstrations and recurrent neural networks to predict robot motion, an approach he relates to today’s imitation learning and world models. Examples include folding clothes, controlling a vehicle and two arms, and teaching a robot to cook scrambled eggs. Multimodal attention helped divide the cooking task into phases so the robot could reproduce skills that are difficult to explain explicitly.

3. 9:39

   Building a shared robotics data and model ecosystem

   Ogata says a Japanese association established the previous December is working toward an open foundation model and dataset. It has collected data across multiple robots and has more than 30 member companies; the plan is for companies to adapt models to their applications and contribute practical data. The association is also developing around-the-clock teleoperation data collection, with operators in different countries.

4. 13:31

   Research priorities and industry deployment

   A separate education ministry-backed project selected 15 proposals focused on fundamental questions, including continuous learning, soft bodies, touch, and sound. Ogata says the industry-focused effort aims to deliver open models, datasets, and tools within three to four years. He calls for collaboration between the research groups and partners beyond Japan.