首页|Researchers at Ben-Gurion University of the Negev Have Published New Data on Rob otics (Automatic Curriculum Determination for Deep Reinforcement Learning in Rec onfigurable Robots)
Researchers at Ben-Gurion University of the Negev Have Published New Data on Rob otics (Automatic Curriculum Determination for Deep Reinforcement Learning in Rec onfigurable Robots)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-Fresh data on robotics are presented i n a new report. According to news originating from Beer-Sheva, Israel, by NewsRx editors, the research stated, "Deep reinforcement learning (DRL) is a prevalent learning method in robotics. DRL is commonly applied in real-world scenarios, s uch as learning motion behavior in rough terrain." Funders for this research include Helmsley Charitable Trust Through The Agricult ural, Biological, And Cognitive Robotics Center Ofben-gurion University. The news correspondents obtained a quote from the research from Ben-Gurion Unive rsity of the Negev: "However, the lengthy learning epochs reduce DRL practicabil ity in many such environments. Curriculum learning can significantly enhance the efficiency of DRL, but establishing a curriculum is challenging, partly because it can be difficult to assess the operation complexity for each task. Determini ng operation complexity can be especially difficult for reconfigurable search an d rescue robots. We present a method for learning based on an automatically esta blished curriculum tuned to the robot's perspective. The method is especially su itable for outdoor environments with multiple obstacle variants, e.g., environme nts encountered in search and rescue missions. After an initial learning stage, the behavior of a robot when overcoming each obstacle variant is characterized u sing Gaussian mixture models (GMMs). Hellinger's distance between the GMMs is co mputed and used to cluster the variants hierarchically. The curriculum is determ ined based on the formed clusters and the average success rate in each cluster. The method was implemented on RSTAR, a highly maneuverable and reconfigurable fi eld robot that can overcome a variety of obstacles. Learning using the automatic ally determined curriculum was compared to learning without a curriculum in a si mulation with three obstacle types: a narrow channel, a low entrance, and a step ."
Ben-Gurion University of the NegevBeer -ShevaIsraelAsiaEmerging TechnologiesMachine LearningNano-robotReinf orcement LearningRobotRobotics
NETL
NSTL
