Active Neural Localization

Abstract

Localization is the problem of estimating the location of an autonomous agent from an observation and a map of the environment. Traditional methods of localization, which filter the belief based on the observations, are sub-optimal in the number of steps required, as they do not decide the actions taken by the agent. We propose "Active Neural Localizer", a fully differentiable neural network that learns to localize accurately and efficiently. The proposed model incorporates ideas of traditional filtering-based localization methods, by using a structured belief of the state with multiplicative interactions to propagate belief, and combines it with a policy model to localize accurately while minimizing the number of steps required for localization. Active Neural Localizer is trained end-to-end with reinforcement learning. We use a variety of simulation environments for our experiments which include random 2D mazes, random mazes in the Doom game engine and a photo-realistic environment in the Unreal game engine. The results on the 2D environments show the effectiveness of the learned policy in an idealistic setting while results on the 3D environments demonstrate the model's capability of learning the policy and perceptual model jointly from raw-pixel based RGB observations. We also show that a model trained on random textures in the Doom environment generalizes well to a photo-realistic office space environment in the Unreal engine.

Under Review at ICLR-18, 15 pages, 7 figures

References in corpus (7)

• High-Dimensional Continuous Control Using Generalized Advantage Estimation
• DeepVO: Towards End-to-End Visual Odometry with Deep Recurrent Convolutional Neural Networks
• Learning to Navigate in Complex Environments
• Domain Randomization for Transferring Deep Neural Networks from Simulation to the Real World
• Neural Map: Structured Memory for Deep Reinforcement Learning
• VidLoc: A Deep Spatio-Temporal Model for 6-DoF Video-Clip Relocalization
• Cosine Normalization: Using Cosine Similarity Instead of Dot Product in Neural Networks

Cited by in corpus (13)

• Neural SLAM: Learning to Explore with External Memory
• Learning to Navigate in Cities Without a Map
• Particle Filter Networks with Application to Visual Localization
• Scene Memory Transformer for Embodied Agents in Long-Horizon Tasks
• Deep Network Uncertainty Maps for Indoor Navigation
• Learning to Visually Navigate in Photorealistic Environments Without any Supervision
• MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation
• Reinforcement Learning-Based Coverage Path Planning with Implicit Cellular Decomposition
• Maximum Reward Formulation In Reinforcement Learning
• Gated Path Planning Networks
• gradSLAM: Automagically differentiable SLAM
• Incremental Scene Synthesis
• Variational State-Space Models for Localisation and Dense 3D Mapping in 6 DoF