LiftNav: Path Planning via Semantic Lifting in TSDF-Guided Gaussian Splatting

2026-05-29

AI Summary

This work addresses the challenge of simultaneously achieving geometrically precise obstacle avoidance and rich object-level semantic understanding in unknown indoor environments. The authors propose a dual-map representation framework that integrates Truncated Signed Distance Fields (TSDF) with Gaussian splatting, incorporating YOLO-based object detection, a TSDF-guided lightweight 3D semantic lifting mechanism, and B-spline trajectory optimization. To enhance trajectory safety and smoothness, a hinge-loss-based collision penalty term is introduced. Notably, the method operates without requiring dense 3D embeddings and achieves 100% path feasibility in the Replica simulation environment, generating navigation trajectories significantly shorter than those produced by current state-of-the-art radiance field approaches.

Abstract

Autonomous robots in unknown indoor environments require both reliable collision avoidance and object-level understanding. Classical representations such as TSDF support safe planning but lack semantics, while photorealistic methods like Gaussian Splatting (GS) provide rich appearance yet suffer from soft geometry, limiting precise obstacle avoidance. We present LiftNav, a hybrid navigation framework built on GSFusion's TSDF+GS dual map, augmented with a real-time pipeline of YOLO-based detection, TSDF-based 3D lifting, and B-spline trajectory optimization. This design enables flexible semantic navigation without dense 3D embeddings. We further introduce a hinge-loss-based collision penalty that improves trajectory smoothness and safety. We evaluate our approach in a simulation using the Replica dataset. Compared against a state-of-the-art radiance field baseline we show a 100% feasibility rate and shorter trajectories.

Problem

• autonomous navigation
• semantic understanding
• collision avoidance
• 3D scene representation
• indoor robotics

Innovation

• Semantic Lifting
• TSDF-Guided Gaussian Splatting
• Hybrid Navigation
• Hinge-Loss Collision Penalty
• Real-Time 3D Semantic Mapping