IROS 2025 Keynotes - Medical Robots: Kanako Harada

In this talk

1. 0:13

   From lab automation to scientific discovery

   Harada argues that AI for science needs more than existing data: researchers must also decide what to collect and how. Robots can interact with the physical world and run experiments to check hypotheses. She distinguishes systems built for efficient, consistent data collection from AI robot scientists that pursue discoveries through their own approaches.

2. 2:57

   Three capabilities for robot scientists

   Harada describes three desired capabilities: generating hypotheses and manipulating tools precisely, detecting subtle changes and targeting samples accurately, and adapting to unfamiliar or hazardous settings. The goal is for robots to carry out experiments that challenge human skill or physical limits, then use results to test scientific ideas.

3. 5:34

   AI-guided experiments and sample collection

   In a plant drug discovery project, AI proposed chemical compounds, robotic tools measured their effects on cell membranes, and AI updated its hypotheses from the results. Harada reports that AI robots found two effective candidates among seven, while scientists found two or four among 1,000; she also describes targeted collection of individual plant root cells and tracking blood cells for unusual behavior.

4. 8:38

   Robotics, simulation, and delicate procedures

   Because life science samples are valuable, Harada says robots must adapt without wasting specimens during training. Her team uses digital twins to train systems for delicate work, including skull drilling on mice with differing skull shapes, and is automating the procedure based on task completion. A virtual interface also lets researchers teach robots to work in confined environments and collaborate across institutions.

5. 13:22

   Toward experiments in space

   Harada describes work with JAXA on a robotic platform for scientific experiments aboard the International Space Station, including a system designed for mouse dissections. She argues that enabling robot scientists to make discoveries beyond repeating human procedures requires collaboration across research disciplines.