To determine whether the person needs medical aid, robots have to recognize vital signs of that person. We propose a method for detecting fallen person by using thermography and point cloud data of three-dimensional SOKUIKI sensor. And also we propose a method for detecting breath of fallen person by using three-dimensional SOKUIKI sensor. In the detection of fallen person, 3D point cloud of a person is extracted based on the detection of body temperature. The position and orientation of the body are estimated from the person's 3D point cloud. For the detection of breath of fallen person, we made preliminary experiment to confirm that breathing can be measured from the cross-sectional shape of the torso. In this method, it is detected whether person is breathing by calculating power spectrum of SOKUIKI sensor data. To realize whole process of fallen person detection and breath detection, we implemented a motion planning of a mobile robot. The robot moves to the position of the body by using proposing fallen person detection, and detects the person's breathing.
A multiple robot environment with decentralized system requires moving obstacle avoidance for robot navigation to prevent collisions between robots. There are many obstacle avoidance methods developed, but most of them are focusing on getting to the goal within the shortest route. In reality, it is necessary to define the accessible area of the robot in a complex indoors environment which may consist of door entrance, descending stairways and others. It is undesirable for a robot to move into these areas where collision is hard to detect or where there is no actual foothold. A method to closely follow the global designated path by setting subgoals and path boundary manually is described. The local path algorithm is based on Tsubouchi et al's method. This method whereby only collision in the front of the robot path is avoided and the robot moves only with the same velocity is enhanced by making the robot move in a range of velocities and thus being able to avoid obstacle staying closely to the designated path. To allow the robot detect obstacle from not only front but also back and sideways, an obstacle detection method using 2 laser range sensors is presented.