Self-valuing learning and generalization with application in visually guided grasping of complex objects
✍ Scribed by Jianwei Zhang; Bernd Rössler
- Publisher
- Elsevier Science
- Year
- 2004
- Tongue
- English
- Weight
- 367 KB
- Volume
- 47
- Category
- Article
- ISSN
- 0921-8890
No coin nor oath required. For personal study only.
✦ Synopsis
For programming by demonstration and for reinforcement learning, the learned skills need appropriate representations for generalization and possibilities for further improvements by the robot itself. We present a self-valuing learning technique which is capable of learning how to grasp unfamiliar objects and generalize the learned abilities. The learning system consists of two components which distinguish between local and global quality criteria for grasp points. The local criteria are not object-specific while the global criteria cover physical properties of each object. In this case we present a generalization method of the learning parameters based on a tree distance model for the medial axis transformations. The system is self-valuing, i.e. it rates its actions by evaluating sensory information and the usage of image processing techniques. This learning system has been implemented in a real robot assembly system equipped with hand-cameras and force/torque sensors. Both the theory and the experiments have shown it ability to grasp a wide range of objects and to apply pre-learned knowledge to new objects.