Computer Vision News 6 3rd AI for Robotics Workshop The third edition of this workshop was held on November 15th and 16th at the NAVER LABS Europe site in Meylan, France. The talks given by invited speakers explored how advances in computer vision, machine learning, scene understanding, natural language processing and related fields will make it possible to equip robots with AI components so they can operate and interact with us in the real world and become integrated in our everyday lives. Huge thanks to Christian Wolf and the Naver Labs Europe team for this awesome report! by the Naver Labs Europe team Inductive biases and structured representations Central questions covered during the workshop were the current challenges in robot learning, in particular the generation and usage of large-scale data required for modern learning algorithms, learning in simulation and the corresponding sim2real gap. Jan Peters, Professor at TU Darmstadt, suggested a whole list of inductive biases which can enable successful learning including the creation of modular policies, incrementally increasing task complexity (curriculum learning), exploiting constraints in exploration and using robot dynamics during learning. Jan Peters (Technische Universität Darmstadt) on how robot learning can benefit from inductive biases. In the same vein of exploiting inductive biases to improve the sample efficiency of the learning process, Sylvain Calignon, senior researcher at IDIAP Research Institute, emphasized the benefits of leveraging the structures of both the data and the geometry when addressing complex tasks in robotics.
RkJQdWJsaXNoZXIy NTc3NzU=