NTR: Neural Token Reconstruction for Scene Token Bottleneck in End-to-End Driving
NTR: Neural Token Reconstruction for Scene Token Bottleneck in End-to-End Driving
This research proposes NTR, a representation learning framework that improves the compact visual representation learning and planning performance in perception-free end-to-end driving by directly constraining the scene-token bottleneck through self-distillation masked latent reconstruction and semantic priors. The method achieves state-of-the-art results on three public autonomous driving benchmarks.
This paper introduces NTR, a novel representation learning framework that addresses the limitations of scene tokens in perception-free end-to-end driving by directly constraining the compact scene-token bottleneck through self-distillation and semantic priors. The method achieves state-of-the-art results on multiple benchmarks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
RFS
8.0461
View original evidence
tion heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim1&2. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervisio
Open source S5PDMS
94.1
View original evidence
tion heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim1&2. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervisio
Open source S5EPDMS
90.9
View original evidence
tion heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim1&2. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervisio
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
tion heads. All auxiliary reconstruction components are removed at inference time, leaving the deployed planner unchanged. NTR achieves state-of-the-art performance on three public autonomous driving benchmarks, including 8.0461 RFS on Waymo E2E and 94.1 PDMS / 90.9 EPDMS on NavSim1&2. The learned scene tokens exhibit lower pairwise redundancy and higher effective rank, indicating that effective bottleneck supervisio
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org