SnapNav: Learning Mapless Visual Navigation with Sparse Directional Guidance and Visual Reference

Learning-based visual navigation still remains a challenging problem in robotics, with two overarching issues: how to transfer the learnt policy to unseen scenarios, and how to deploy the system on real robots. In this paper, we propose a deep neural network based visual navigation system, SnapNav. Unlike map-based navigation or Visual-Teach-and-Repeat (VT&R), SnapNav only receives a few snapshots of the environment combined with directional guidance to allow it to execute the navigation task. Additionally, SnapNav can be easily deployed on real robots due to a two-level hierarchy: a high level commander that provides directional commands and a low level controller that provides real-time control and obstacle avoidance. This also allows us to effectively use simulated and real data to train the different layers of the hierarchy, facilitating robust control. Extensive experimental results show that SnapNav achieves a highly autonomous navigation ability compared to baseline models, enabling sparse, map-less navigation in previously unseen environments.

Authors

• Linhai Xie
• Andrew Markham
• Niki Trigoni

Keywords

• Robots
• Navigation
• Visualization
• Task analysis
• Training
• Collision avoidance
• Turning
• Machine Vision
• Visual Reference
• Deep Neural Network
• Obstacle Avoidance
• Low-level Control
• Real Robot
• Problem In Robotics
• Layers Of Hierarchy
• Convolutional Neural Network
• Generalization Ability
• Small Datasets
• Recurrent Network
• Raw Images
• Virtual World
• Positive Image
• Depth Images
• Optical Flow
• Current Observations
• Linear Velocity
• Policy Learning
• Left Turn
• Metric Learning
• Policy Gradient Method
• Real-world Test
• Unknown Environment
• Robot Dynamics
• Linear Embedding
• Attention Layer
• Reward Function
• Pseudo Labels