Problem
Research questions and friction points this paper is trying to address.
End-to-End Autonomous Driving
Trajectory Planning
Multi-Scale Semantic Features
Multi-Camera Fusion
S2Planner: Multi-Scale Semantic Planner for End-to-End Autonomous Driving
🤖 AI Summary
This study addresses the insufficient integration of multi-scale semantic and geometric constraints in end-to-end trajectory planning for autonomous driving by proposing a planning framework that fuses visual inputs, ego-vehicle history, and navigational commands. Methodologically, it employs a DINOv3 backbone combined with spatial tuning adapters to extract multi-scale features, and designs an ego-conditioned trajectory initialization scheme alongside a geometry-guided multi-scale feature sampling mechanism. The trajectory waypoints are subsequently iteratively refined through a coarse-to-fine decoder. Evaluated on the NAVSIM v1 benchmark, the proposed model achieves a PDMS of 88.03, validating the effectiveness of the architecture. Nevertheless, its generalization capability and computational efficiency warrant further assessment.
📝 Abstract
We present S2Planner, a trajectory planner that combines three front-facing cameras with ego-motion history and the current driving command. A fine-tuned DINOv3 backbone and a Spatial Tuning Adapter produce multi-scale image features; a coarse-to-fine decoder then uses trajectory self-attention and camera-projected cross-attention to refine candidate waypoints. The contribution is the integration of ego-conditioned trajectory initialization with iterative, geometry-guided sampling of multi-scale image features, rather than a new visual backbone or attention operator. On the NAVSIM v1 non-reactive evaluation, the previously reported navtest run obtained 88.03 PDMS. Because that run was selected using navtest performance, this number is exploratory and cannot be interpreted as an unbiased test estimate. Validation-selected evaluation on unexposed data, repeated runs, and computational measurements are needed to establish generalization and efficiency.
Innovation
Methods, ideas, or system contributions that make the work stand out.
End-to-End Autonomous Driving
Multi-Scale Semantic Planner
Trajectory Planning
Geometry-Guided Sampling
Ego-Conditioned Initialization
🔎 Similar Papers
No similar papers found.
