In recent years, artificial intelligence has become increasingly important for understanding the real world and the innate human ability to process intuitive physics with computers. Most research on real-world detection and prediction has used methods that generate predictive images of the environment from changes in the pixels of an image, or that predict changes in objects in the environment from changes in the numerical values of a physical simulator. However, the actual method of predicting the environment in humans is believed to consist of both visual and physical information. Therefore, in this study, in order to recognize the motion of objects, the relationship of motion characteristics between objects is represented by a graph structure, which enables the extraction of anomalous points in response to changes in the environment. Then, based on both visual and physical information, we constructed a change point prediction model that can predict the motion of objects in the environment and capture the timing of collisions between objects. Moreover, to improve the accuracy, the basic prediction model was modified and the accuracy of the model was compared. The results showed that visual information and physical information were inferred independently, and the more physical information included the physical properties of the objects, the higher the accuracy. In the change point prediction model, as the prediction accuracy of the base prediction model increased, the change point extraction accuracy also increased.
20 September 2025 in Journal of Advanced Computational Intelligence and Intelligent Informatics
In recent years, prediction models for the real world have been widely proposed. Most research that deals with the recognition and prediction of the real world generates prediction results from visual predictions such as changes in pixels or numerical changes in physical simulators, and few models can predict them based on both visual and physical characteristics, as humans can. Therefore, in this study, we constructed a new prediction model based on both visual information and physical characteristics in the environment by integrating the mechanism of variational temporal abstraction, which extracts change points in the observation environment from visual information, into PreCNet. Furthermore, to make the prediction results interpretable, we generated the inferred prediction content as a sentence. In addition, we verified whether the generated sentences could explain collision situations in as much detail as a human being when given physical common sense about the environment, such as the movement and mass of objects.