Acquisition of Dynamics Matching in Sensory-Motor Fusion System
Summary
As sensory-motor fusion systems work faster, the system performance becomes more susceptible to constraints derived from physical and computational dynamic properties of the systems. We proposed a novel concept named "dynamics matching" which aims the maximization of the system performance by adjusting of the temporal properties (dynamics) of sensors, processors and actuators, and making them consistent with dynamics of tasks or environments under physical and computational constraints.
In this study, we proposed a model of dynamics matching problems for sensory-motor fusion systems, and built an algorithm for adaptive acquisition of dynamics matching using reinforcement learning.
Some numerical experiments of target-tracking task by an active vision using the proposed algorithm illustrated the adequate adjustment of dynamics according to various constraints, and the acquisition of rational solution in the behavior-level.
References
- Naoko Ogawa, Yutaka Sakaguchi, Akio Namiki and Masatoshi Ishikawa. Adaptive Acquisition of Dynamics Matching in Sensory-Motor Fusion System. Electronics and Communications in Japan (Part III: Fundamental Electronic Science), Vol. 89, No. 7, pp. 19-30, Jul. 2006. [PDF (1.0M)] [doi:10.1002/ecjc.20255] *Wiley)
- Naoko Ogawa, Yutaka Sakaguchi, Akio Namiki and Masatoshi Ishikawa. Adaptive Dynamics Matching in Sensory-Motor Fusion System and Its Implementation by Reinforcement Learning. 2002 IEEE International Conference on Robotics and Automation (ICRA 2002) (Washington D.C., USA, 2002.5.11) / In Learning Skill for Enhancing the Capability of Artificial System (Y. Sakaguchi ed.), Workshop Note, May 2002. [PDF (120K)]
