Active Information Gathering for Long-Horizon Navigation Under Uncertainty by Learning the Value of Information

TL;DR

This work tackles long-horizon point-goal navigation under uncertainty in partially mapped environments by inferring the long-horizon value of information $V_I$ for exploratory actions and integrating it into a Learning over Subgoals (LSP) planning framework. A Graph Neural Network estimates subgoal properties and the value of information, enabling the planner to actively seek information that improves planning performance while maintaining completeness and soundness. The approach computes one-step information gains $v_I$ and accumulates them as $V_I$ during offline training to train the estimator, and then applies these predictions at deployment time. In three simulated office-like environments, the method significantly reduces average navigation cost compared with non-learned baselines and prior LSP variants, with improvements up to 63.76% and full goal-reachability, demonstrating practical benefits of principled information-seeking for long-horizon navigation under uncertainty.

Abstract

We address the task of long-horizon navigation in partially mapped environments for which active gathering of information about faraway unseen space is essential for good behavior. We present a novel planning strategy that, at training time, affords tractable computation of the value of information associated with revealing potentially informative regions of unseen space, data used to train a graph neural network to predict the goodness of temporally-extended exploratory actions. Our learning-augmented model-based planning approach predicts the expected value of information of revealing unseen space and is capable of using these predictions to actively seek information and so improve long-horizon navigation. Across two simulated office-like environments, our planner outperforms competitive learned and non-learned baseline navigation strategies, achieving improvements of up to 63.76% and 36.68%, demonstrating its capacity to actively seek performance-critical information.

Figures (7)

• Figure 1: Overview: actively gathering information is essential for good navigation in a partial map. Baseline approach reaches the goal slowly, encountering many dead-ends, while our approach (LSP-AIG) actively gathers useful information that lets it quickly reach the goal.
• Figure 2: Low cost navigation in our J-Intersection environment requires active gathering of information. When the goal is either on left or right from the intersection, knowing the information contained at the center of the map allows us to decide correctly at the intersection. Choosing always left or right or even choosing one color over another will not reliably succeed.
• Figure 3: Our robot's actions correspond to boundaries between free and unseen space. The robot can leave observed space through either boundary: via subgoal $s_1$ or $s_2$. Upon selecting action $a_2$, the robot reaches the goal with probability $P_S$ and incurs an expected cost $R_S$, or is turned back (probability $1-P_S$), accumulates cost $R_E$ and selects another action.
• Figure 4: Value of information training data example. We show the total value of information $V_I$ (cumulatively summed over the one step value of information $v_I$) for the action that contains the map information. Color signifies time step in both plots, enabling easy visual correspondence between the two.
• Figure 5: Ring Office environment Results: scatter-plots and example trials. Our LSP-AIG planner outperforms both the non-learned baseline and the LSP-GNN planners in 100 trials by actively gathering the information required to efficiently navigate in our Ring Office environment.