Keynotes

In reinforcement learning, an agent learns behaviors to maximize expected cumulative rewards. An open issue in reinforcement learning applications is how to design a reward function for a desired behavior. A related issue in neuroscience is what rewards really are in animals and humans. This talk reviews computational models of intrinsic rewards to promote exploration, inverse reinforcement learning methods to infer a reward function from the behavior, and embodied evolution experiments to test whether robots can acquire their own reward functions for survival and reproduction.

Medicine-Engineering research presents unique challenges, particularly in quantifying medical needs and evaluating the performance of medical devices. To mitigate this, our team has developed 'Bionic Humanoids'—sensor-embedded patient models that streamline device development and provide a balanced comparison of procedures with and without these devices.
Building on this, we've developed various surgical robots, including a transnasal neurosurgical robot, a paediatric surgical robot, and an ophthalmic surgical robot. We've demonstrated that the use of bionic humanoids accelerates Medicine-Engineering research. The future of surgical robotics leans towards autonomy—a realm filled with both challenges and opportunities. We will also introduce our pioneering efforts towards autonomous surgical robots and our attempts to leverage this technology in scientific experiments.