Robot Sequential Decision Making using LSTM-based Learning and Logical-probabilistic Reasoning.

摘要: Sequential decision-making (SDM) plays a key role in intelligent robotics, and can be realized very different ways, such as supervised learning, automated reasoning, probabilistic planning. The three families of methods follow assumptions have (dis)advantages. In this work, we aim at robot SDM framework that exploits the complementary features We utilize long short-term memory (LSTM), for passive state estimation with streaming sensor data, commonsense reasoning planning (CORPP) active information collection task accomplishment. experiments, mobile is tasked estimating human intentions using their motion trajectories, declarative contextual knowledge, human-robot interaction (dialog-based motion-based). Results suggest our performs better than its no-learning no-reasoning versions real-world office environment.