Gpu-accelerated collision-aware planning improves 3d gaussian splatting scenes
CollisionSplatting: Collision-Aware Motion Planning in 3DGS Scenes with Image-Conditioned Objectives and Adjustable Conservatism
Robotics
Summary
Planning safe movement for robots using detailed 3D scenes is hard because traditional methods simplify visuals or are slow and hard to interpret. The authors propose a new technique called CollisionSplatting that uses a fast, adjustable way to measure distances directly on detailed 3D visual data. This method works well with AI tools that use images to guide decisions and fits nicely into popular robot path planning methods. Their approach runs fast, uses less memory, and makes smart, safe navigation possible in real-world tasks.
What this means in practice
- •For robot navigation teams: Enable real-time, collision-aware path planning using rich visual 3D data while saving memory and computing resources.
- •For robotic manipulation developers: Guide robotic arms in cluttered environments using integrated visual and geometric collision metrics for safer handling.
Authors
R. Khorrambakht, Joaquim Ortiz-Haro, Stephan Weiss, Ludovic Righetti
Abstract
Incorporating dense visual information into motion planning remains challenging, as geometric planners rely on abstracted scene representations that discard visual richness, while learned visual models often lack geometric interpretability and computational efficiency. This paper introduces CollisionSplatting, a simple, modular, GPU-accelerated, probability-inspired distance metric with tunable conservatism that operates directly on standard 3D Gaussian Splatting (3DGS) scenes. When combined with learned image-conditioned reward functions, this metric enables joint geometric and visual planning by unifying collision-aware costs with image-space objectives. We integrate the metric into GPU-accelerated Model Predictive Path Integral (MPPI) and Rapidly-Exploring Random Tree (RRT) planners, and show on-par or better collision-classification performance compared to representative baselines while achieving substantially higher collision-checking throughput and significantly lower VRAM usage. Finally, we demonstrate the effectiveness of our metric in real-world vision-guided navigation and manipulation tasks, highlighting 3DGS as a practical bridge between rich perception and real-time motion planning.