Digital Twin-Assisted Robot Navigation Using RF Ray-Tracing: A Privacy-Preserving Approach
This paper introduces DT-RaDaR, a novel robot navigation framework that prioritizes user privacy by leveraging RF ray-tracing within a digital twin environment, effectively circumventing the need for privacy-invasive sensors like cameras or LiDAR.
HEIGHT: Using a Heterogeneous Interaction Graph Transformer to Improve Robot Navigation in Cluttered and Crowded Environments
This paper introduces HEIGHT, a novel robot navigation policy network that leverages a heterogeneous spatio-temporal graph to model interactions between robots, humans, and obstacles in crowded and constrained environments, leading to safer and more efficient navigation compared to existing methods.
FlowNav: Using Conditional Flow Matching for Efficient Robot Navigation
This research paper introduces FlowNav, a novel approach for robot navigation that leverages Conditional Flow Matching (CFM) to generate efficient and accurate action policies, outperforming traditional diffusion-based methods in terms of speed and achieving comparable accuracy.
Hybrid and Oriented Harmonic Potentials for Safe Robot Task Execution in Unknown Environments
This paper proposes a novel method for safe and efficient robot navigation in unknown environments by combining harmonic potential fields for local navigation with oriented search trees for global task planning.
Navigation with Uncertainty-driven Exploration (NUE): Enhancing Visual Navigation in Robots Using NeRF Uncertainty for Exploration
This research introduces NUE, a novel robotic visual navigation system that leverages the uncertainty inherent in online-trained NeRF models to drive exploration, leading to more efficient learning of unknown environments and improved image-goal navigation performance.
NeuronsGym: 시뮬레이션과 실제 환경 학습을 위한 하이브리드 프레임워크 및 벤치마크
NeuronsGym은 로봇 내비게이션 정책 학습 및 시뮬레이션 환경에서 실제 환경으로의 전이 능력을 평가하기 위한 하이브리드 프레임워크 및 벤치마크를 제시하며, 향상된 안전성 평가 지표인 SFPL을 소개하고 다양한 시뮬레이션-실제 환경 전이 방법들의 성능을 비교 분석합니다.
FRTree Planner: A Novel Approach to Real-Time Robot Navigation in Cluttered and Unknown Environments
The FRTree Planner enables real-time, collision-free robot navigation in cluttered and unknown environments by dynamically constructing a tree of free regions and using a bi-level trajectory optimization that considers the robot's geometry.
X-MOBILITY: An End-to-End Generalizable Navigation Model Using World Modeling for Improved Performance in Robotics
X-MOBILITY, a novel end-to-end navigation model, leverages world modeling and imitation learning to achieve superior performance and generalization capabilities in robot navigation tasks, outperforming existing methods in challenging, out-of-distribution environments.
Active Neural Time Fields for Efficient Robot Mapping and Motion Planning in Unknown Environments
This paper introduces Active NTFields, a novel framework that leverages physics-informed neural networks to efficiently map unknown environments and generate real-time motion plans for robots by learning arrival time fields directly from sensor data.
OrionNav: A Framework for Real-Time Robot Navigation in Unknown and Dynamic Environments Using Open-Vocabulary Semantic Scene Graphs and a Context-Aware Large Language Model
OrionNav enables robots to autonomously navigate unknown, dynamic environments and perform diverse tasks by combining real-time open-vocabulary semantic mapping, hierarchical scene graph generation, and LLM-based planning with low-level motion control.
© 2024 by Linnk AI