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

Hannah Schieber, Dominik Frischmann, Victor Schaack, Angela P. Schoellig, Daniel Roth

Keywords: Gaussian Splatting, Navigation, Semantics, Path Planning

Abstract: Autonomous robots operating in unknown indoor environments require both reliable collision avoidance and flexible object-level understanding. Classical geometric representations such as TSDF enable safe planning but lack semantic expressiveness, while photorealistic representations like Gaussian Splatting (GS) provide rich visual cues yet suffer from “soft geometry,” limiting their use for precise obstacle avoidance. We present LiftNav, a hybrid navigation framework built on GSFusion’s TSDF+GS dual map, augmented with our real-time pipeline: YOLO-based detection, TSDF-based 3D lifting, and B-spline trajectory optimization, enabling flexible semantic navigation without dense 3D embeddings. Additionally, we introduce a hinge-loss-based collision penalty that strongly improves trajectory smoothness and safety. Extensive evaluation on the Replica dataset against a state-of-the-art radiance field baseline shows a 100\% feasibility rate, shorter trajectories, and up to a 4x reduction in maximum jerk, demonstrating robust, agile, and semantically-aware navigation.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 46