Early-Signal Detector

method · first seen 5 Oct 2026

end-to-end autonomous driving

Also: E2E driving, end-to-end driving, end-to-end navigation

Driving policies that map sensor inputs directly to vehicle control or trajectories in one learned model.

End-to-end autonomous driving trains a single model to map camera or LiDAR input directly to trajectories or controls, replacing hand-built perception, prediction and planning modules. The sources are a steady stream of incremental work rather than a new idea. They include query-based cost learning over reachable trajectories, sparse alternatives to dense BEV grids (BEVCast, LOWP), JEPA-style and latent world-action models, risk-sensitive candidate ranking, cooperative intent sharing, and benchmarks for in-context decisions. OpenDriveLab released code studying whether visual foundation models help planning, and one team adapted openpilot for NVIDIA's AlpaSim closed-loop challenge. The active subtopics are world models, VLA planners and closed-loop evaluation.

The overall paradigm has been a mainstream research direction since UniAD and earlier imitation-learning work, so this rise mostly reflects normal paper volume. The subthreads worth watching are latent world models and closed-loop evaluation in neural-reconstructed scenes.

Analyst note (established), 17 Nov 2026.

Mentions per week

0819 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 163 on 14 Nov 2026 · 18 mentions in 28 days · 16 independent groups · papers → code · first seen 6 weeks ago · acceleration +1.22

Earliest sources

Latest sources

Mentioned together with

flow matchingworld modelbird's-eye-view representationvision-language-action modeljoint-embedding predictive architecturemodel predictive path integral controldiffusion policytrajectory planningANN-to-SNN conversionspiking neural networks