Learning to Explore using Active Neural SLAM

Abstract

This work presents a modular and hierarchical approach to learn policies for exploring 3D environments, called `Active Neural SLAM'. Our approach leverages the strengths of both classical and learning-based methods, by using analytical path planners with learned SLAM module, and global and local policies. The use of learning provides flexibility with respect to input modalities (in the SLAM module), leverages structural regularities of the world (in global policies), and provides robustness to errors in state estimation (in local policies). Such use of learning within each module retains its benefits, while at the same time, hierarchical decomposition and modular training allow us to sidestep the high sample complexities associated with training end-to-end policies. Our experiments in visually and physically realistic simulated 3D environments demonstrate the effectiveness of our approach over past learning and geometry-based approaches. The proposed model can also be easily transferred to the PointGoal task and was the winning entry of the CVPR 2019 Habitat PointGoal Navigation Challenge.

Published in ICLR-2020. See the project webpage at https://devendrachaplot.github.io/projects/Neural-SLAM for supplementary videos. The code is available at https://github.com/devendrachaplot/Neural-SLAM

Cited by in corpus (22)

• Object Goal Navigation using Goal-Oriented Semantic Exploration
• AllenAct: A Framework for Embodied AI Research
• Causal Navigation by Continuous-time Neural Networks
• Collaborative Visual Navigation
• Move to See Better: Self-Improving Embodied Object Detection
• Deep Learning for Embodied Vision Navigation: A Survey
• Topological Planning with Transformers for Vision-and-Language Navigation
• The ThreeDWorld Transport Challenge: A Visually Guided Task-and-Motion Planning Benchmark for Physically Realistic Embodied AI
• MVP: Unified Motion and Visual Self-Supervised Learning for Large-Scale Robotic Navigation
• No RL, No Simulation: Learning to Navigate without Navigating
• Unsupervised Domain Adaptation for Visual Navigation
• SASRA: Semantically-aware Spatio-temporal Reasoning Agent for Vision-and-Language Navigation in Continuous Environments
• Shaping embodied agent behavior with activity-context priors from egocentric video
• Beyond Tabula-Rasa: a Modular Reinforcement Learning Approach for Physically Embedded 3D Sokoban
• Learning to Navigate Sidewalks in Outdoor Environments
• MaAST: Map Attention with Semantic Transformersfor Efficient Visual Navigation
• Semantic Audio-Visual Navigation
• Learning Synthetic to Real Transfer for Localization and Navigational Tasks
• Learning Composable Behavior Embeddings for Long-horizon Visual Navigation
• Building Intelligent Autonomous Navigation Agents
• Realistic PointGoal Navigation via Auxiliary Losses and Information Bottleneck
• Graph Convolutional Memory using Topological Priors