Advances in Embodied Intelligence and Navigation

The field of embodied intelligence and navigation is rapidly advancing, with a focus on developing more efficient and robust methods for navigating complex environments. Recent research has explored the use of hierarchical structures, semantic-aware approaches, and adaptive curriculum learning to improve navigation performance. Notably, the use of bigraphs and open scene graphs has shown promise in organizing and maintaining spatial information effectively at scale. Additionally, the integration of curiosity-driven exploration and intrinsic rewards has led to more robust and diverse exploration strategies. The development of benchmarks such as UAV-ON has also facilitated the evaluation and comparison of navigation methods in open-world environments. Some noteworthy papers include: GeoExplorer, which proposes a curiosity-driven exploration approach for active geo-localization, and UAV-ON, which introduces a benchmark for open-world object goal navigation with aerial agents. SA-GCS, which presents a semantic-aware Gaussian curriculum scheduling framework for UAV vision-language navigation, and GACL, which proposes a grounded adaptive curriculum learning approach for robotics tasks. CogniPlan, which leverages conditional generative layout prediction for uncertainty-guided path planning, and SkeNa, which introduces a sketch map-based visual navigation task and a large-scale dataset for research. $NavA^3$, which proposes a hierarchical framework for understanding high-level human instructions and performing spatial-aware object navigation, and Open Scene Graphs, which introduces a modular system for open-world object-goal navigation using foundation models and open scene graphs. HDDPG, which proposes a hierarchical deep deterministic policy gradient algorithm for autonomous maze navigation of mobile robots.

Sources

Building Bigraphs of the real world

GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration

UAV-ON: A Benchmark for Open-World Object Goal Navigation with Aerial Agents

SA-GCS: Semantic-Aware Gaussian Curriculum Scheduling for UAV Vision-Language Navigation

GACL: Grounded Adaptive Curriculum Learning with Active Task and Performance Monitoring

CogniPlan: Uncertainty-Guided Path Planning with Conditional Generative Layout Prediction

SkeNa: Learning to Navigate Unseen Environments Based on Abstract Hand-Drawn Maps

$NavA^3$: Understanding Any Instruction, Navigating Anywhere, Finding Anything

Open Scene Graphs for Open-World Object-Goal Navigation

Hierarchical Deep Deterministic Policy Gradient for Autonomous Maze Navigation of Mobile Robots

Built with on top of