Research Of Indoor Mobile Robots Semantic Target Navigation Method For Visual Perception

The autonomous navigation is the important challenge of robot production.Traditional robotic depends on the scene model,and the accuracy of model has a great influence the performance of navigation,so it is difficult for robots to make decision in complex environment with great generalization.With the development of artificial intelligence,the environmental perception ability has improved a lot,and provides a new concept for the intelligence of robot navigation.Compare with traditional navigation intelligence,robots can take the semantic navigation in known or unknown scene by giving the special target and find the target object.Based on the challenge of the scene model's precision and navigation's generalization,this paper propose two framework for the different semantic navigation scene of application.For the semantic navigation in known scene,the representation and parsing are very important for the understanding of robots.However,parsing and utilizing of the 3D scene is not trivial due to the complexity of unstructured 3D space and the limited reasoning ability of the robot.This study propose a framework for structured 3D scene graph generation,which efficiently describes the objects,relations and sttributes of the 3D indoor environment with structured representation.In the proposed method,we adopt visual perception to capture the semantic information and inference from scene priors to calculate the optimal parse graph.Afterwards,an improved probabilistic grammar model is used to represent the scene priors.Experiment results demostrate that the peoposed framework significantly outperform existing methods in terms of accuracy,and in applying to high-level human-robot interaction tasks.In the unknown scene,the ability of making decision for visual target navigation is still the core challenge.With the development of deep reinforcement learning,more and more researcher try to use the end-to-end method to deal with this task.However,duo to the difficulty in learning navigational policy and complicated representation of the scene menory and priors,the performance and efficiency of the semantic navigaiton in unknown scene is non-trivial for autonomous robots.Hence,we introduce a new deep reinforcement learning framework based on the secene priors and semantic map.In the peoposed framework,the semantic map is built as scene memory and the 3D scene knowledge is encoded by relational graph convolution network.Based on the real-time map and explicit scene priors,the navigational policy is learned to sample the goal by deep reinforcement learning.Experiment results demonstratr how semantic map and scene priors improve performance significantly in family indoor scene.