NeOR: Neural Exploration with Feature-based Visual Odometry and Tracking-Failure-Reduction Policy Author Name Affiliation Postcode Ziheng Zhu  Zhejiang University of Technology  310023 Jialing Liu  Zhejiang University of Technology  Kaiqi Chen  Zhejiang University of Technology  Qiyi Tong  Zhejiang University of Technology  Ruyu Liu*  Hangzhou Normal University  311121 Abstract:       Embodied visual exploration is critical for building intelligent visual agents. This paper presents NeOR, a framework for embodied visual exploration that possesses the efficient exploration capabilities of deep reinforcement learn-ing-based exploration policies and leverages feature-based visual odometry (VO) for more accurate mapping and positioning results. An improved local policy is also proposed to reduce tracking failures of feature-based VO in weakly textured scenes through a refined multi-discrete action space, keyframe fusion, and an auxiliary task. The experimental results demonstrate that NeOR has better mapping and positioning accuracy compared to other en-tirely learning-based exploration frameworks and improves the robustness of feature-based VO by significantly reducing tracking failures in weakly textured scenes. Hits: 0 Download times: 0 The National Key Technologies R&D Program of China;The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan);China Postdoctoral Science Foundation     Download reader