Abstract:By adopting evolution algorithm and autonomous-vision-based fitness evaluation approaches,the on-line au- tomatic gait evolution of 4-legged robot in a RoboCup soccer field is realized.We incorporate interpolation method as the crossover method,use a PC base station to conduct algorithm calculation and flow control,and adopt some time-cutting strategies.The evolutionary learning is implemented with high continuity and expansibility,and the whole learning process can be completed within 40-60 minutes.In-field gait re-learning of the ERS-7 4-legged robot is realized,and the adaptability of walking control is improved.At last,the walking speed of ERS-7 4-legged robots is increased from 27 cm/s to 43 cm/s with the proposed algorithm.
[1] Hengst B,Ibbotson D,Pham S B,et al.The UNSW RoboCup 2000 Sony legged league team[A].RoboCup 2000:Robot Soccer World Cup Ⅳ[C].Berlin,Germany:Springer,2001.64~75.
[2] Mitchell T M.Machine Learning[M].USA:McGraw-Hill,1997.
[3] Powell M J D.A method for nonlinear constraints in minimization problems[A].Optimization[M].London,UK:Academic Press,1969.283~298.
[4] Baird L,Moore A W.Gradient descent for general reinforcement learning[J].Advances in Neural Information Processing Systems,1999,11:968~974.
[5] Kim M S,Uther W.Automatic gait optimization for quadruped robots[A].Proceedings of the Australasian Conference on Robotics and Automation[C].Brisbane,Australia:Australian Robotics and Automation Association,2003.
[6] Kohl N,Stone E Policy gradient reinforcement learning for fast quadrupedal locomotion[A].Proceedings of the IEEE International Conference on Robotics and Automation[C].Piscataway NJ,USA:IEEE,2004.2619~2624.
[7] Zhang J Q,Chen Q J.Lamarckian evolution and Baldwin effect in neural networks[J].DCDIS A Supplement,Advances in Neural Networks,2007,14(S1):470~473.
[8] Back T.Evolutionary Algorithms in Theory and Practice:Evolution Strategies,Evolutionary Programming,Genetic Algorithms[M].New York,USA:Oxford University Press,1996.
[9] Sony Corporation.Model information for ERS-7[EB/OL].http://www.robots.newcastle.edu.au/QuinlanThesis/ModelInformation_7_E.pdf,2004.
[10] Duffert U.Quadruped walking modeling and optimization of robot movements[EB/OL] ,http://uwe-dueffert.de/publication/dueffert04_diploma.pdf,2004-07-22.
[11] Gutmann J S.Markov-Kalman localization for mobile robots[A].Proceedings of the 16th International Conference on Pattern Recognition[C].Piscataway,NJ,USA:IEEE,2002.601~604.
[12] Quinlan M J,Nicklin S P,Hong K,et al.Nubot final report 2005[EB/OL].http://robots.newcastle.edu.au/publications/NUbotFinalReport2005.pdf,2006-02-02.
[13] Chen Q J,Xie H L,Woo P Y.Vision-based fast objects recognition and distances calculation of robots[A].31st Annual Conference of IEEE Industrial Electronics Society[C].Piscataway,NJ,USA:IEEE,2005.363~368.