This paper proposes a selective merging method of 2D range scans for enhancing our CIF-based scan matching algorithm which is a global scan matching algorithm using the CIF descriptors and a geometric constraint between keypoints. The selective merging method can generate a local map based on “terrain complexity” without fixing the number of scans to use. Our enhanced version of the CIF-based scan matching algorithm can perform global scan matching more robustly in a large cluttered environments without using an initial alignment. Through experiment in real environment, we confirm the validity of our method.

Our research purpose is to estublish an automated manipulator system which can operate a deformable linear object. Especially, our research group has been focusing on the string shape operation by a robot. We devide process of shape operation into four phases. These are string shape reconstruction, planning of string shape operation, robot motion generation and execution of shape operation. In our previous study on string shape reconstruction, shape of a string was reconstructed from 3D point cloud data by proposed method. In this study, the planning method of string untying operation based on knot theory and the motion generation of a manipulator are proposed. Finally, the results of string untying experiment are reported.

The recognition of indicated objects by interacting people is an essential function for robots that act in daily environments. However, due to ambiguous references by them, accurate recognition of indicated objects have difficulties for the robots. For example, people sometimes use the words which did not contain in the robots' databases, or they did not use enough words to identify the object. Therefore, to improve recognition accuracy, we must decrease such ambiguity of indicating behaviors of people. For this purpose, we experimentally compared two kinds of interaction strategies to decrease the ambiguity: explicitly providing requests to people about how to refer to objects, or implicitly aligning with people's indicating behaviors. The experimental results showed that participants evaluated the implicit strategy to be more natural than the explicit strategy, and the recognition performances of the two strategies were not significantly different.