GUIDEd Agents: Enhancing Navigation Policies through Task-Specific Uncertainty Abstraction in Localization-Limited Environments

TL;DR

This work introduces Task-Specific Uncertainty Maps (TSUMs) and the GUIDE framework to enable navigation policies that selectively allocate localization resources based on task context. By conditioning policies on per-location uncertainty budgets derived from task relevance, constraints, and environment, GUIDE enables agents to balance task progress with uncertainty management without reward engineering. The authors implement a GUIDEd Soft Actor-Critic (G-SAC) and demonstrate improved task completion and efficiency in real-world autonomous surface vehicle experiments compared to baselines that ignore task-specific uncertainty. The approach addresses localization-limited navigation challenges and provides a path toward more resource-aware robotic planning in complex environments.

Abstract

Autonomous vehicles performing navigation tasks in complex environments face significant challenges due to uncertainty in state estimation. In many scenarios, such as stealth operations or resource-constrained settings, accessing high-precision localization comes at a significant cost, forcing robots to rely primarily on less precise state estimates. Our key observation is that different tasks require varying levels of precision in different regions: a robot navigating a crowded space might need precise localization near obstacles but can operate effectively with less precision elsewhere. In this paper, we present a planning method for integrating task-specific uncertainty requirements directly into navigation policies. We introduce Task-Specific Uncertainty Maps (TSUMs), which abstract the acceptable levels of state estimation uncertainty across different regions. TSUMs align task requirements and environmental features using a shared representation space, generated via a domain-adapted encoder. Using TSUMs, we propose Generalized Uncertainty Integration for Decision-Making and Execution (GUIDE), a policy conditioning framework that incorporates these uncertainty requirements into robot decision-making. We find that TSUMs provide an effective way to abstract task-specific uncertainty requirements, and conditioning policies on TSUMs enables the robot to reason about the context-dependent value of certainty and adapt its behavior accordingly. We show how integrating GUIDE into reinforcement learning frameworks allows the agent to learn navigation policies that effectively balance task completion and uncertainty management without explicit reward engineering. We evaluate GUIDE on various real-world robotic navigation tasks and find that it demonstrates significant improvement in task completion rates compared to baseline methods that do not explicitly consider task-specific uncertainty.

Figures (10)

• Figure 1: Top: The ASV is assigned a navigation task. GUIDE interprets the task and generates a representation highlighting areas where the ASV needs higher positional certainty (dark blue). Bottom: The policy executed by the ASV (white line). The red dots indicate locations where the ASV actively reduces its state estimation uncertainty to satisfy task-specific requirements.
• Figure 2: GUIDE consists of two phases. (i) Pre-training: semantic and spatial embeddings are fine-tuned and aligned using a CLIP-based model. (ii) Deployment: TSUMs are generated from task descriptions and policies are conditioned on the TSUMs to manage uncertainty.
• Figure 3: CLIP-based TSUM generation pipeline. Environment imagery and task description are encoded by a fine-tuned CLIP; their joint embeddings produce task-relevance and constraint maps that, fused with environment priors, yield the final TSUM.
• Figure : GUIDEd SAC
• Figure : GUIDEd SAC