The main point of this work is to provide a process for learning dynamics models from a single stack of demo videos to a fully operational model without any network blowing. In this work, we proposed the concepts, processes and all the pipelines needed to refine, the correct model from a dynamic model based on any person demo videos. Our findings revealed, demonstrations are in some ways a low cost format for collecting data and can be used to learn dynamics models. In addition, this data format can be used to a greater number of domains (dialogues between training programs) than can be difficult to gather in classical training procedures. On the basis of that, this work can be done, showcase of one of the aspect of socio-technical systems management, which is the human-machine interface and interaction, where it is the subject of a sector that has developed an industry in recent years and is growing rapidly. We introduce here a method for environmental control systems of which, the methods established in this work, can be applied directly to other areas.
Robots are currently being used for a wide range of applications due to their optimality in the manipulation of situations which are difficult to control (e.g. rapid change in position). The social interactive robots can allow us to interact with the environment, mainly connected to a high degree of efficiency in dynamic situations. In general, three categories of mechanisms are applied in order to emulate the real behavior of humans in human-robot interactions, as follows: (i) imitation learning, (ii) model-based learning, and (iii) knowledge-based learning. But, this work focuses on the behavior of interaction patterns, and then regards the case of robot interaction modeling. For this reason, the model-based learning is the method should be considered.
This paper is an initial investigation on the use of a combination of model-based learning and model selection for the environmental control. This approach builds upon the previous work on RL and learning of model-based method with demonstration videos. Here, we experimented to use the data collected from a single human demonstrations. Specifically, we focus on continuous control of robotic arm. Incorporated to the best of our knowledge, the current work is the first attempt to model the interactive control and behavior in an open environment, as well as an important research contribution. d2c66b5586