Artificial intelligence must still fill the gap “Sim-to-Rél”. In -depth learning techniques that are all the rage in the superlative performance of the LOG in the mastery of brain games, including failures and go, which can both be played on a computer. But the translation of simulations in the physical world remains a greater challenge.
A robot named Curly who uses “a deep reinforcement learning” – providing improvements because she corrects her own mistakes – went out in three of the four games against the leading human opponents of the South Korean teams whoIncluded a female team And a reserve team for the national team of wheelchairs. (No brooms have been used).
A crucial conclusion is that the AI system demonstrated its ability to adapt to changing ice conditions. “These results indicate that the gap between simulators based on physics and the real world can be narrowed,” wrote the joint South Korean-German research team in Scientific robotics September 23.