A wide range of accurate environmental information is needed in order for autonomous mobile robots to move safely. An intelligent space, which contains multiple sensors attached to the environment, is useful for information provision. Many optical sensors such as Laser Range Finders (LRFs) are placed to monitor fixed areas. But, LRFs occasionally miss objects due to distance attenuation of lasers, irregular reflection of lasers, and gaps between lasers. Thus, possibility of these detection failures of LRFs should be considered. In this paper, a novel measurement model with properties of the two-dimensional LRFs is constructed, and the empty probability distribution created by the model are proposed. We set two-dimensional LRFs up on a roadside, and calculated an empty probability distribution to evaluate the utility of the proposed model. In addition, since intelligent spaces often contain multiple LRFs, data measured from different viewpoints can be fused to improve data accuracy. We also evaluated sensor-fusion of multiple empty probability distributions obtained from individual LRFs. The fusion result grasps a location at which a detection failure of LRF has occurred. The usefulness of the proposed method was confirmed in our experimental evaluation.
This paper proposes a force control design method for complex multi-step tasks. Our proposed method is composed of two processes: a human operator performs the same task by operating a master-slave robotic system, and then the parameter extracted from human demonstration is further optimized to minimize the cycle time of the task. In our proposed method, the force control parameters are extracted from the human demonstration data by using a hidden Markov model. To determine the parameter switching condition, a simple experiment is performed. Finally, an optimization to minimize the cycle time is carried out. The parameters extracted from human demonstration data are used as initial values, and a novel optimization algorithm that is specialized for cycle time minimization is applied. An experiment of insert a rubber packing into the groove of the inner surface of a pipe part was conducted. We show that the cycle time was reduced to almost half during the optimization process.
This paper presents a motion planning method for a bipedal robot to walk through terrain constructed by various slopes. The method uses terrain information obtained from proximity sensors which are mounted on each foot. The obtained information changes especially due to a motion of the swing leg because the proximity sensor detects a local terrain. Hence, the proposed method generates the trajectory every control cycle by using the terrain information obtained by the proximity sensor which has fast responsiveness (< 1[ms]). In our previous work, a design method was proposed for an arc-shaped proximity sensor detecting two relative amounts (distance and posture) between the sensor and object independently. We also presented a motion planning method for bipedal walking on even terrain, which can reduce energy consumption by a rolling effect of the arc-shaped foot. The proposed method in this paper uses the above method on even terrain and a trajectory generation method using the sensor if the tilt is changed. The collaboration of two motion planning methods achieves both lower energy consumption on even terrain and walking through the tilt changing terrain. In two types of simulation, the trajectory generation adapted to random slopes and a walking the above features are confirmed.