Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning-based Methods for Adaptive Informative Path Planning

Marija Popovic, Joshua Ott, Julius Rückin, Mykel J. Kochenderfer

TL;DR

This survey tackles adaptive informative path planning (AIPP) in robotics, focusing on how learning-based methods can enhance online data gathering in unknown environments. It presents a unified mathematical framework for AIPP, and offers two complementary taxonomies: one by learning approaches (supervised, RL, imitation, active learning) and one by robotic applications (environmental monitoring, exploration, semantic understanding, and active SLAM). The paper analyzes mapping models (GPs, occupancy grids, graphs, and implicit neural representations), evaluation metrics, and benchmarks, and discusses challenges around generalization, localization uncertainty, temporal dynamics, heterogeneity, and the lack of standardized evaluation. By linking AIPP to active and imitation learning and outlining future directions, it highlights how learning-based AIPP can yield more robust, scalable, and versatile data-gathering robotics systems with open-source resources and benchmarks. The work thus provides a reference roadmap for researchers aiming to develop generalizable, uncertainty-aware, and benchmarked learning-enabled AIPP solutions for real-world deployments.

Abstract

Adaptive informative path planning (AIPP) is important to many robotics applications, enabling mobile robots to efficiently collect useful data about initially unknown environments. In addition, learning-based methods are increasingly used in robotics to enhance adaptability, versatility, and robustness across diverse and complex tasks. Our survey explores research on applying robotic learning to AIPP, bridging the gap between these two research fields. We begin by providing a unified mathematical framework for general AIPP problems. Next, we establish two complementary taxonomies of current work from the perspectives of (i) learning algorithms and (ii) robotic applications. We explore synergies, recent trends, and highlight the benefits of learning-based methods in AIPP frameworks. Finally, we discuss key challenges and promising future directions to enable more generally applicable and robust robotic data-gathering systems through learning. We provide a comprehensive catalogue of papers reviewed in our survey, including publicly available repositories, to facilitate future studies in the field.

Learning-based Methods for Adaptive Informative Path Planning

TL;DR

This survey tackles adaptive informative path planning (AIPP) in robotics, focusing on how learning-based methods can enhance online data gathering in unknown environments. It presents a unified mathematical framework for AIPP, and offers two complementary taxonomies: one by learning approaches (supervised, RL, imitation, active learning) and one by robotic applications (environmental monitoring, exploration, semantic understanding, and active SLAM). The paper analyzes mapping models (GPs, occupancy grids, graphs, and implicit neural representations), evaluation metrics, and benchmarks, and discusses challenges around generalization, localization uncertainty, temporal dynamics, heterogeneity, and the lack of standardized evaluation. By linking AIPP to active and imitation learning and outlining future directions, it highlights how learning-based AIPP can yield more robust, scalable, and versatile data-gathering robotics systems with open-source resources and benchmarks. The work thus provides a reference roadmap for researchers aiming to develop generalizable, uncertainty-aware, and benchmarked learning-enabled AIPP solutions for real-world deployments.

Abstract

Adaptive informative path planning (AIPP) is important to many robotics applications, enabling mobile robots to efficiently collect useful data about initially unknown environments. In addition, learning-based methods are increasingly used in robotics to enhance adaptability, versatility, and robustness across diverse and complex tasks. Our survey explores research on applying robotic learning to AIPP, bridging the gap between these two research fields. We begin by providing a unified mathematical framework for general AIPP problems. Next, we establish two complementary taxonomies of current work from the perspectives of (i) learning algorithms and (ii) robotic applications. We explore synergies, recent trends, and highlight the benefits of learning-based methods in AIPP frameworks. Finally, we discuss key challenges and promising future directions to enable more generally applicable and robust robotic data-gathering systems through learning. We provide a comprehensive catalogue of papers reviewed in our survey, including publicly available repositories, to facilitate future studies in the field.
Paper Structure (24 sections, 5 equations, 3 figures, 5 tables)

This paper contains 24 sections, 5 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Mathematical Formulation
  3. Background
  4. Mapping
  5. Evaluation Metrics
  6. Benchmarks
  7. Simulations
  8. Common and Classical Baselines
  9. Learning Approaches
  10. Supervised Learning
  11. Reinforcement Learning
  12. Network Architectures and Training Algorithms
  13. Reinforcement Learning Approaches
  14. Imitation Learning
  15. Active Learning
  16. ...and 9 more sections

Figures (3)

  • Figure 1: Overview of our survey paper. We review related work in learning-based methods for adaptive informative path planning (AIPP) from the perspectives of learning-based approaches (left) and practical applications (right).
  • Figure 2: System diagram showing the key elements of a general AIPP framework. During a mission, a map $\tilde{F}$ of the robot's environment $\xi$ is built using measurements $z$ extracted from a sensor data stream. An AIPP algorithm leverages the map data to find the next action sequences $\psi$ for the robot to maximize the information value ${I}(\psi)$, i.e. utility, of future measurements. The next action sequence $\psi$ is executed by the robot, allowing for subsequent map updates in a closed-loop manner.
  • Figure 3: Summary statistics for the papers reviewed in \ref{['S:applications']}.