Learning to Navigate in Complex Environments

Abstract

Learning to navigate in complex environments with dynamic elements is an important milestone in developing AI agents. In this work we formulate the navigation question as a reinforcement learning problem and show that data efficiency and task performance can be dramatically improved by relying on additional auxiliary tasks leveraging multimodal sensory inputs. In particular we consider jointly learning the goal-driven reinforcement learning problem with auxiliary depth prediction and loop closure classification tasks. This approach can learn to navigate from raw sensory input in complicated 3D mazes, approaching human-level performance even under conditions where the goal location changes frequently. We provide detailed analysis of the agent behaviour, its ability to localise, and its network activity dynamics, showing that the agent implicitly learns key navigation abilities.

11 pages, 5 appendix pages, 11 figures, 3 tables, under review as a conference paper at ICLR 2017

Cited by in corpus (38)

• A Brief Survey of Deep Reinforcement Learning
• Towards Monocular Vision based Obstacle Avoidance through Deep Reinforcement Learning
• Self-Monitoring Navigation Agent via Auxiliary Progress Estimation
• DeepMDP: Learning Continuous Latent Space Models for Representation Learning
• Unifying Map and Landmark Based Representations for Visual Navigation
• The StreetLearn Environment and Dataset
• Benchmarking Classic and Learned Navigation in Complex 3D Environments
• Search on the Replay Buffer: Bridging Planning and Reinforcement Learning
• Learning Exploration Policies for Navigation
• Active Neural Localization
• Visual pathways from the perspective of cost functions and multi-task deep neural networks
• Observational Learning by Reinforcement Learning
• Learning Robotic Manipulation of Granular Media
• Discriminative Particle Filter Reinforcement Learning for Complex Partial Observations
• Scene Memory Transformer for Embodied Agents in Long-Horizon Tasks
• The Regretful Agent: Heuristic-Aided Navigation through Progress Estimation
• How You Act Tells a Lot: Privacy-Leakage Attack on Deep Reinforcement Learning
• Learning to Navigate in Indoor Environments: from Memorizing to Reasoning
• What Should I Do Now? Marrying Reinforcement Learning and Symbolic Planning
• Are You Looking? Grounding to Multiple Modalities in Vision-and-Language Navigation
• Tactical Rewind: Self-Correction via Backtracking in Vision-and-Language Navigation
• Motion Perception in Reinforcement Learning with Dynamic Objects
• Learning to Move with Affordance Maps
• Graph Attention Memory for Visual Navigation
• Stratospheric Aerosol Injection as a Deep Reinforcement Learning Problem
• Deep execution monitor for robot assistive tasks
• Towards continuous control of flippers for a multi-terrain robot using deep reinforcement learning
• Learning Object-conditioned Exploration using Distributed Soft Actor Critic
• Active 6D Multi-Object Pose Estimation in Cluttered Scenarios with Deep Reinforcement Learning
• Active Visual Information Gathering for Vision-Language Navigation
• Meta-Learning to Cluster
• LeRoP: A Learning-Based Modular Robot Photography Framework
• Learning to Look Around: Intelligently Exploring Unseen Environments for Unknown Tasks
• Learning Autonomous Exploration and Mapping with Semantic Vision
• Vision-based deep execution monitoring
• Implementing Inductive bias for different navigation tasks through diverse RNN attractors
• Partially Observable Planning and Learning for Systems with Non-Uniform Dynamics
• MAPEL: Multi-Agent Pursuer-Evader Learning using Situation Report