China
Africa
Explore chapters and articles related to this topic
Big data-based deterioration prediction models and infrastructure management: towards assetmetrics
Published in Dan M. Frangopol, Hitoshi Furuta, Mitsuyoshi Akiyama, Dan M. Frangopol, Life-Cycle of Structural Systems, 2018
Kiyoshi Kobayashi, Kiyoyuki Kaito
A similar hierarchical relation can be found between the crack rate and the asphalt load-bearing capacity. The asphalt load-bearing capacity can be evaluated by means of falling weight deflectometer (FWD) tests. However, FWD tests are not performed regularly, but only when necessary. Moreover, an FWD test requires a temporary traffic regulation. Thus, road administrators cannot perform them frequently. Moreover, it is extremely important to make good decisions regarding where these should be conducted. The hierarchical hidden Markov model is a model that expresses the hierarchical relation between the crack rate and the asphalt load-bearing capacity (i.e. the implicit knowledge that when the asphalt load-bearing capacity decreases, the crack rate rapidly increases). Figure 4 depicts an example of the prediction result. The degradation processes for the crack rate and asphalt load-bearing capacity are modelled using a five-step Markov degradation hazard model. The continuous red line in the figure corresponds to the expected degradation path for the asphalt load-bearing capacity. It can be seen that the expected life is about 50 years.
Hierarchical and parameterized learning of pick-and-place manipulation from under-specified human demonstrations
Published in Advanced Robotics, 2020
Kun Qian, Huan Liu, Jaime Valls Miro, Xingshuo Jing, Bo Zhou
The task decomposition process typically involves a segment-recognition-composition framework. As an example, Konidaris [28] proposes to construct skill trees from trajectories, in which a sequence of skill segments, each with a short term goal, is learned from human demonstrated trajectories. For multiple trajectories of repeated demonstration, a complete skill tree is constructed by merging multiple skill chains. The author conducted task sequence verification experiments on the uBot-5 manipulator. Nevertheless, such a segment-recognition-composition framework only takes into account trajectories, without the consideration of object manipulations. Probabilistic models such as Hierarchical Hidden Markov Model (HHMM) [29] are capable of representing behaviors at multiple levels of abstraction. They are beneficial for enhancing the generalization performance of the robot's reproduced behaviors. Patel [30] proposes to use HHMM to segment the path of a mobile robot in its daily task. With a three-layer dynamic Bayesian network model, the motion primitives and task labels are inferred in a bottom-up fashion based on the observation sequences of the robot's own states. Sanzari [31] proposed the concept of motion flux to represent the behavior characteristics of human joint movements in a short time. The unknown motion primitives were discovered by optimizing the motion flux from the 3D human body posture in the video sequence, and the unknown primitives were labeled unsupervisedly by the non-parametric Bayesian mixed model.