In a closed-door discussion that I attended recently, the award committee was debating whether to award the best student paper to a reinforcement learning-based robotics work with simulations using open datasets, or a robotics work using more traditional model-based methods with real robot experiments. Much more deliberation took place to reach the decision, and I’m not going to reveal which paper won the award, although, I must say both papers deserve recognition for being finalists in the competition. As a researcher who studied both classic dynamic model-based methods and machine learning-based methods to control robots, I have no bias toward either one, and I think the best solution for complex robotics problems may lie in better integrating of the two methods.
Footer
