Скачать с ютуб Trained Probabilistic Models for the NAO Robot in a Labyrinth в хорошем качестве

Trained Probabilistic Models for the NAO Robot in a Labyrinth

A framework based on a Probabilistic Model for the moving behavior of the NAO humanoid robot in the environment given by $20in \times 20in$ vinyl maze cells is being trained and made available for future applications. NAO is one of the most advanced humanoid robots, having advanced speech, vision, and behavior based on artificial intelligence already implemented on it, and being a precursory of the larger Pepper robot famous for being used as host at certain hotels. Pepper uses wheels, probably since the leg-based movement of Nao proved hard to harness with precision and robustness. Indeed, most of the intelligence currently present in NAO is speech and gesture-related, while its autonomous walking capabilities are only little exploited in existing available software, and only with reflexes without high-level utility-driven intelligence. We test that it is possible to exploit a public NAO sensor database made recently available, to build a sample probabilistic model for walking and turning in a controlled vinyl maze. The probabilistic model is a new and powerful representation of related phenomena based on random variables and with conditional probability tables for the NAO sensors computed from experimental measurements given relevant environment states. The model allows for complex planners and reasoners, that are based on rich POMDP models, to be built on top of it. Such a high-level AI framework allows for easily giving NAO new tasks by just specifying the corresponding utilities. The proposed model is tested with a simple particle filter localizer on a predefined trajectory, and improvements and data missing in the Nao database are being identified for future work.