This paper describes a graph search-based local path planner with an adaptive node sampling according to positions of obstacles. Randomly sampled nodes of a graph in a traversable region for finding a local path can generate a winding path due to connection between the randomly sampling nodes. Node sampling with constant intervals can fail to find a proper path in a case that an interval between adjacent obstacles are smaller than the sampling interval. To solve these problems, our path planner changes node sampling strategy according to obstacle positions; it samples nodes along a reference path, e.g., the center line of the lane, if there are no obstacles, but densely samples nodes around obstacles. After sampling, the planner searches the graph using Dijkstra's algorithm for finding an optimal trajectory. For efficient search for the optimal trajectory, the planner firstly generates a trajectory approximated by piecewise linear lines with minimum cost, and fine-tunes it by adjusting node positions with curved edges having smaller cost. We test the planner through simulations in which an ego vehicle drive along a reference path when there are no obstacles, and densely sample around obstacles to improve robustness against obstacle locations.

This paper proposes a method that reasons intent of motion through task order such as "switch on the light." Intent of motion means kinds of path of a contact point between a target and an end effector when a robot recreates ordered task. The intent is used for teaching. After a robot fixes a framework of task using the intent, it asks questions to a user to specify concrete values in the framework, the user replies them. This interaction creates motions of the robot. The kinds of path are defined as linear, circular and point to point. Moreover, impossible task is also considered. Four sets of P/sub f/(w): Probability that a word represents a path, and C/sub p/(w}: Certainty of P/sub p/(w) are added to each word in an electric thesaurus to enable the reasoning. After values are input in part of the sets by 8 production rules, they are propagated to the other sets through hierarchical relationship of words in the thesaurus. When the user enters two words, namely task and target, argmax/sub p/{P/sub p/(w)*C/sub p/(w)} becomes answer. Moreover, update and re-propagation of values in the sets using answers from the user make the reasoning more precise.

The design for an artificial mark attached to small objects is presented in this paper. Three conditions are required for the design of the marks: (1) short calculation time, (2) easy attachment, and (3) individual calculation for each item. The marks are obtained by using three different extraction methods. The experiments determine the best mark. The mark using the method that extract colors and calculated the gravity has good results. We analyze the errors of the mark that has best experiment. The appropriateness of the experimental results is then confirmed. We have the experiment that the manipulator handles the objects with the best mark. The experiment is successful. Therefore, the usability of the mark is verified.

To achieve smooth motion of car-like robots, it is necessary to generate paths that satisfy the following conditions: maximum curvature, maximum curvature derivative, and curvature continuity. Another requirement is that human operators can manipulate the robots with ease. In this paper, a path expression methodology consisting of line segments, circular arcs and clothoid arcs is presented. In addition, a method of global path generation with a visibility graph is proposed. To establish this method, the following steps are proposed: (a) the arrangement of subgoals (middle points) and (b) the construction of the graph for path generation. By using the proposed method, the paths were shortened 14% on average.

This paper presents a robot system for handling various objects in home or office environments. A fixed manipulator utilizes marks on objects for handling and motion planning. A mark consists of two parts. One is the outer part, which indicates its pose (in this paper, "pose " denotes 3D position and orientation), and the other is a QR code, which is a kind of 2D barcode. A QR code is stored information of an object (e.g. its name). A user attaches several marks on each object. The manipulator accesses the information using a camera in its hand. The robot can decide its complex handling motion based on the information in QR codes, pose of the objects, which are estimated from pose of marks, and signals from proximity sensors. Experiments are conducted to verify the whole system.

Recently the activity of robots has been remarkable, and robots are branching out into general environment such as home or office. Although robots are active, planners uniquely decide the action method of robots when robots operate missions. Therefore, the most suitable approach is not assured. So, in this report we propose a method that we evaluate robots' actions from the viewpoint of both reliability and time, when robots operate missions. Then we calculate the most suitable approach for robots. To be concrete, by using a diagram we evaluate robots' actions and calculate the most suitable approach. In addition, we evaluate them not from all the actions immediately, but from the action units by making an evaluation index. So, we can shorten the time for evaluation. Finally, we realize the operations by a real robot with the most appropriate approach.