The article addresses topics about 3D geometric fitting for multiple objects and describes novel hemi-form geometric models for single-scan point clouds and probabilistic formulation with a combination of Gaussian mixture models and EM algorithm. The hemi-form models assign larger residuals to the observations locating behind a geometric model so as to reflect the fact that those observations in single-scan data are unlikely to belong to that model. Since our models such as hemisphere, hemi-cylinder, and hemi-cone models, allow the range of residuals to be symmetric, we can assume Gaussian distributions as a probabilistic model of the residuals in the strict sense. Calculation of geometric model parameters is implemented by a nonlinear least-squares method and initial value estimation techniques. The proposed framework was applied to both synthetic and real world point clouds involving multiple objects as well as outliers and was found to return acceptable model estimation accuracy results.
This paper proposes a power assist unit using an add-on effector so as to easily construct a power assist system. In the proposed system, the assisting force of the actuator is transmitted to the environment via a pair of assist springs. Thus, the desired assisting force can be generated by controlling the spring displacement in accordance with the operator-applying force. The controller is designed taking into consideration explicitly both the human and environmental mechanical impedances in order to analyze the stability for the parameter variations and to make the system more robust against those perturbations. The system also allows the operator to adjust the amplitude of assisting ratio, i.e., the assist gain, online. This function could improve usability and widen its application field such as rehabilitation and welfare equipment. Performance of a prototype of the proposed system was evaluated by experiments in both time and frequency domain for different back drivability conditions. Experimental results showed that the prototype has robustly worked against perturbations of the environment, keeping the power assist control bandwidth approximately 1 to 3 Hz, and the validity of the variable assist gain control has also confirmed.
This paper discusses a dynamic nonprehensile manipulation of a deformable object, where the shape of a deformable object is controlled by using the plate's rapid vibration. After experimentally confirming the feasibility of the manipulation principle, we introduce a simplified analytical model where a deformable object is modeled by two mass points and the plate has two degrees of freedom: a translational motion and a rotational one. Using this model, we investigate how the object's behavior changes with respect to the amplitude of the rotational angular acceleration of the plate. We show that the behaviors of the mass points and the whole object are categorized by the six non-dimensional boundary amplitudes. Through simulation analysis, we then reveal that the deformation velocity transition of the object is characterized by the boundary amplitudes. We make clear that the optimal plate's motion leading to the maximal deformation velocity is provided by one of the six boundary amplitudes.
The human understanding of things is based on prediction which is made through concepts formed by categorization of their experience. To mimic this mechanism in robots, multimodal categorization, which enables the robot to form concepts, has been studied. On the other hand, segmentation and categorization of human motions have also been studied to recognize and predict future motions. This paper addresses the issue of how these concepts are integrated to generate higher level concepts and, more importantly, how the higher level concepts affect each lower level concept formation. To this end, we propose multi-layered multimodal latent Dirichlet allocation (mMLDA) to learn and represent the hierarchical structure of concepts. We also examine a simple integration model and compare with the mMLDA. The experimental results reveal that the mMLDA leads to better inference performance and, indeed, forms higher level concepts integrating motions and objects that are necessary for real-world understanding.