Deep Learning for Embodied Vision Navigation: A Survey

Abstract

The problem of embodied vision navigation attracts rising attention from the community due to its wide application in autonomous driving, vacuum cleaner, and rescue robot. To accomplish embodied vision navigation requires various intelligent skills, such as exploration, mapping, and planning, visual recognition and reasoning, etc. Moreover, building a robot that observes like a human, thinks like a human, and acts like a human helps the community to know how intelligence really is. 3D simulation technology provides a large scale of data to simulate the real-world environment, which enables researchers to train and test the embodied visual navigation model within it. The robustness of deep learning methods empowered the embodied navigation agents to accomplish diverse tasks. However, embodied navigation is still in its infancy due to a lot of challenges, like learning a robust policy from partially observed visual input, learning to exploration in navigation, accomplishing natural language navigation tasks, adapting to the real-world environment, etc. Recently, numerous works have been proposed to tackle different challenges in this area. To give a promising direction for future research, in this paper, we present a comprehensive review of embodied navigation tasks and the recent progress in deep learning-based methods.

Vision-and-Language Navigation: A Survey of Tasks, Methods, and Future Directions

“…An agent in Visual Navigation tasks is usually not required to understand information from textual modality. Visual Navigation (Zhu et al, 2021b) is a problem of navigating an agent from the current location to find the goal target. Researchers have achieved success in both simulated environments Mirowski, 2019) and real environments (Mirowski et al, 2018).…”

Task-Driven Graph Attention for Hierarchical Relational Object Navigation

“…II. RELATED WORK a) Object navigation: Embodied search for objects with partial observability is a long-studied problem in embodied AI and robotics [10,21,22,[24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39][39][40][41]. Over the past several years many works have focused on learningbased approaches, since such approaches require less prior knowledge in novel environments [42].…”

Contrastive Instruction-Trajectory Learning for Vision-Language Navigation

“…Vision-Language Navigation (VLN) task (Anderson et al 2018b) requires an agent to navigate following a natural language instruction. This task is closely connected to many real-world applications, such as household robots and rescue robots (Zhu et al 2021). The VLN task is challenging since it requires an agent to acquire diverse skills, such as vision-language alignment, sequential vision perception and long-term decision making.…”