%0 Journal Article %1 kamio:2005:TEC %A Kamio, Shotaro %A Iba, Hitoshi %D 2005 %J IEEE Transactions on Evolutionary Computation %K (RL) (artificial AIBO Adaptation HOAP-1 Q-learning adaptation adaptive algorithms, box box-moving computation, evolutionary four-legged genetic humanoid intelligence), learning learning, legged locomotion, method, moving, programming, real reinforcement robot, robots, systems, task, technique, %N 3 %P 318--333 %R doi:10.1109/TEVC.2005.850290 %T Adaptation technique for integrating genetic programming and reinforcement learning for real robots %V 9 %X We propose an integrated technique of genetic programming (GP) and reinforcement learning (RL) to enable a real robot to adapt its actions to a real environment. Our technique does not require a precise simulator because learning is achieved through the real robot. In addition, our technique makes it possible for real robots to learn effective actions. Based on this proposed technique, we acquire common programs, using GP, which are applicable to various types of robots. Through this acquired program, we execute RL in a real robot. With our method, the robot can adapt to its own operational characteristics and learn effective actions. In this paper, we show experimental results from two different robots: a four-legged robot AIBO and a humanoid robot HOAP-1. We present results showing that both effectively solved the box-moving task; the end result demonstrates that our proposed technique performs better than the traditional Q-learning method.

@article{kamio:2005:TEC, abstract = {We propose an integrated technique of genetic programming (GP) and reinforcement learning (RL) to enable a real robot to adapt its actions to a real environment. Our technique does not require a precise simulator because learning is achieved through the real robot. In addition, our technique makes it possible for real robots to learn effective actions. Based on this proposed technique, we acquire common programs, using GP, which are applicable to various types of robots. Through this acquired program, we execute RL in a real robot. With our method, the robot can adapt to its own operational characteristics and learn effective actions. In this paper, we show experimental results from two different robots: a four-legged robot AIBO and a humanoid robot HOAP-1. We present results showing that both effectively solved the box-moving task; the end result demonstrates that our proposed technique performs better than the traditional Q-learning method.}, added-at = {2008-06-19T17:35:00.000+0200}, author = {Kamio, Shotaro and Iba, Hitoshi}, biburl = {https://www.bibsonomy.org/bibtex/2a79d72a39fe048d430a23baa51b845be/brazovayeye}, doi = {doi:10.1109/TEVC.2005.850290}, interhash = {ca001209c905485b3ed3719a09a212d6}, intrahash = {a79d72a39fe048d430a23baa51b845be}, issn = {1089-778X}, journal = {IEEE Transactions on Evolutionary Computation}, keywords = {(RL) (artificial AIBO Adaptation HOAP-1 Q-learning adaptation adaptive algorithms, box box-moving computation, evolutionary four-legged genetic humanoid intelligence), learning learning, legged locomotion, method, moving, programming, real reinforcement robot, robots, systems, task, technique,}, month = {June}, notes = {INSPEC Accession Number: 8465512 Graduate Sch. of Frontier Sci., Univ. of Tokyo, Chiba, Japan. Reinforcement learning (RL) is outside the GP loop. Q-table 168 or 238 states, RL ten hours or 6 hours. 6 way IF. GP run ten minutes. Target and box are colour coded. Precise simulation of both robots is not possible. p 330 GP {"}learning some general knowledge...not limited to a particular robot{"}.}, number = 3, pages = {318--333}, size = {16 pages}, timestamp = {2008-06-19T17:42:51.000+0200}, title = {Adaptation technique for integrating genetic programming and reinforcement learning for real robots}, volume = 9, year = 2005 }