Object Recognition Using Tactile Measurements: Kernel Sparse Coding Methods
Object Recognition Using Tactile Measurements: Kernel Sparse Coding Methods
复制标题
DOI：
10.1109/tim.2016.2514779
复制
发表时间：
2016-03-01
影响因子：
5.6
通讯作者：
Sun, Fuchun
中科院分区：
文献类型：
--
作者：
Liu, Huaping;Guo, Di;Sun, Fuchun
在过去的十年中，灵巧的机器人出现了，以满足在外科手术，海底焊接和太空机械操作等各种应用中对精细运动控制辅助的需求。对于精细操作和接触环境感知至关重要的是固定在机器人指尖上的触觉传感器。这些可用于区分材料纹理、粗糙度、空间特征、顺应性和摩擦力。本文将所研究的触觉数据视为时间序列，可以通过流行的动态时间扭曲方法来评估其相异性。因此，核稀疏编码方法的开发，以解决触觉数据的表示和分类问题。然而，稀疏编码的天真使用忽略了同时接触对象的各个手指之间的内在关系。为了解决这个问题，我们开发了一个联合核稀疏编码模型来解决多指触觉序列分类问题。在该模型中，手指之间的内在关系被明确考虑使用联合稀疏编码，这鼓励所有的编码向量共享相同的稀疏支持模式。实验结果表明，联合稀疏编码比传统稀疏编码具有更好的性能。
Dexterous robots have emerged in the last decade in response to the need for fine-motor-control assistance in applications as diverse as surgery, undersea welding, and mechanical manipulation in space. Crucial to the fine operation and contact environmental perception are tactile sensors that are fixed on the robotic fingertips. These can be used to distinguish material texture, roughness, spatial features, compliance, and friction. In this paper, we regard the investigated tactile data as time sequences, of which dissimilarity can be evaluated by the popular dynamic time warping method. A kernel sparse coding method is therefore developed to address the tactile data representation and classification problem. However, the naive use of sparse coding neglects the intrinsic relation between individual fingers, which simultaneously contact the object. To tackle this problem, we develop a joint kernel sparse coding model to solve the multifinger tactile sequence classification problem. In this model, the intrinsic relations between fingers are explicitly taken into account using the joint sparse coding, which encourages all of the coding vectors to share the same sparsity support pattern. The experimental results show that the joint sparse coding achieves better performance than conventional sparse coding.