Fast Navigation Through Occluded Spaces via Language-Conditioned Map Prediction

2025-12-24

AI Summary

To address the safety-efficiency trade-off in cluttered environments caused by occlusions and limited sensor field-of-view, this paper proposes a language-guided hierarchical local motion planning framework. Methodologically, it introduces natural-language co-pilot instructions into both local map prediction and subgoal generation—establishing a two-level neural prediction architecture: Level-1 (semantic-augmented map expansion) and Level-2 (instruction-conditioned subgoal generation), integrated with a Log-MPPI optimal controller. The system jointly processes LiDAR/depth observations and textual instructions to produce interpretable, goal-directed, and robust trajectories. Evaluated in polygon-based simulated environments, the approach achieves a 36% improvement in navigation success rate over pure local-map baselines, while significantly enhancing occlusion traversal efficiency and trajectory stability.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Natural Language Processing: Safety and Robustness
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

In cluttered environments, motion planners often face a trade-off between safety and speed due to uncertainty caused by occlusions and limited sensor range. In this work, we investigate whether co-pilot instructions can help robots plan more decisively while remaining safe. We introduce PaceForecaster, as an approach that incorporates such co-pilot instructions into local planners. PaceForecaster takes the robot's local sensor footprint (Level-1) and the provided co-pilot instructions as input and predicts (i) a forecasted map with all regions visible from Level-1 (Level-2) and (ii) an instruction-conditioned subgoal within Level-2. The subgoal provides the planner with explicit guidance to exploit the forecasted environment in a goal-directed manner. We integrate PaceForecaster with a Log-MPPI controller and demonstrate that using language-conditioned forecasts and goals improves navigation performance by 36% over a local-map-only baseline while in polygonal environments.

Problem

• Enhance robot navigation in cluttered, occluded environments using language instructions
• Predict future map visibility and subgoals from local sensor data and co-pilot commands
• Improve planning decisiveness and safety by integrating forecasted environmental information

Innovation

• Language-conditioned map prediction for navigation
• Forecasts occluded areas using local sensor data
• Generates instruction-based subgoals for efficient planning