공 던지기 로봇의 정책 예측 심층 강화학습
Deep Reinforcement Learning of Ball Throwing Robot’s Policy Prediction
0
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
Robot's throwing control is difficult to accurately calculate because of air resistance and rotational inertia, etc. This complexity can be solved by using machine learning. Reinforcement learning using reward function puts limit on adapting to new environment for robots. Therefore, this paper applied deep reinforcement learning using neural network without reward function. Throwing is evaluated as a success or failure. AI network learns by taking the target position and control policy as input and yielding the evaluation as output. Then, the task is carried out by predicting the success probability according to the target location and control policy and searching the policy with the highest probability. Repeating this task can result in performance improvements as data accumulates. And this model can even predict tasks that were not previously attempted which means it is an universally applicable learning model for any new environment. According to the data results from 520 experiments, this learning model guarantees 75% success rate.
Abstract
Robot's throwing control is difficult to accurately calculate because of air resistance and rotational inertia, etc. This complexity can be solved by using machine learning. Reinforcement learning using reward function puts limit on adapting to new environment for robots. Therefore, this paper applied deep reinforcement learning using neural network without reward function. Throwing is evaluated as a success or failure. AI network learns by taking the target position and control policy as input and yielding the evaluation as output. Then, the task is carried out by predicting the success probability according to the target location and control policy and searching the policy with the highest probability. Repeating this task can result in performance improvements as data accumulates. And this model can even predict tasks that were not previously attempted which means it is an universally applicable learning model for any new environment. According to the data results from 520 experiments, this learning model guarantees 75% success rate.
참고문헌 (13)
- U. Jung, "Design of an Efficient Robotic Arm Model for Accurate Throwing", The Korean Association of Computer Education, 19(2), pp.121~125 (2015)
- E. W. Aboaf, "Task-Level Robot Learning", 1988 IEEE International Conference on Robotics and Automation (1988)
- A. Zeng, "Tossingbot : Learning to Throw Arbitrary Objects with Residual Physics", IEEE Transactions on Robotics, 36(4) (2020)[DOI]
- A. Ghadirzadeh, "Deep Predictive Policy Training using Reinforcement Learning", 2017IEEE, 2017, ~
- J. Kober, "Reinforcement Learning in Robotics : A Survey", The International Journal of Robotics Research, 32(11) (2013)[DOI]
- "Batch Active Preference-Based Learning of Reward Functions, arXiv:1810. 04303" (2018)[URL]
- J. Kober, "Reinforcement Learning to Adjust Robot Movements to New Situations", Autonomous Robots, 33 (2011)[DOI]
- J. Y. Park, "Performance Comparison of Crawling Robots Trained by Reinforcement Learning Methods", Korean Institute of Intelligent Systems, 17(1), pp.33~36 (2007)
- 양영하, "보틀플리핑의 로봇 강화학습을 위한 효과적인 보상 함수의 설계", 로봇학회 논문지, 14(2), pp.081~086 (2019), 한국로봇학회[DOI]
- P. Abbeel, "Using Inaccurate Models in Reinforcement Learning", The 23rd international conference on Machine learning (2006)
- H. Yokota, "Information visualisation of Optimised Underhand Throw for Cybernetic Training", Procedia Engineering, 112 (2015)[DOI]
- J. Kober, "Policy Search for Motor Primitives in Robotics", Springer Tracts in Advanced Robotics, 97 (2009)[DOI]
- S. Jung, "A New Neural Network Control Technique for Robot Manipulators", 1995 American Control Conference-ACC’95 (1995)