Neural Radiance Fields
The implicit ancestor: representing scenes as continuous radiance functions, sampled by ray marching.
-
★ Oral · Best Paper Honorable Mention
The paper that started everything — volume rendering meets MLPs.
-
★ Oral · Honorable Mention
-
★ Oral
-
★ Oral
Sparse voxel grids — the first hint that explicit primitives could beat MLPs on speed.
-
★ Best Paper
Instant-NGP — multi-resolution hash grids; training a NeRF in seconds.
-
★ Highlight
Gaussian Splatting · Autonomous Driving
Explicit radiance primitives applied to street-scale reconstruction — dynamic, unbounded, multi-sensor.
-
★ Best Paper
The paper that opened the door — explicit primitives that train and render in minutes.
-
★ Best Paper Runner-up
-
★ Highlight
-
Unified NeRF backbone for AD; influential baseline for SplatAD.
-
★ Spotlight
-
Direct predecessor to our line of work — camera + lidar Gaussian splatting in driving.
Online HD Map Construction
Building lane and structure maps in the loop, from onboard sensors and external priors. Vector-first; failure-robust.
-
Established the rasterized-output baseline most modern methods compare against.
-
★ Spotlight
Vectorized output via permutation-invariant point sets — the dominant paradigm since.
-
★ Oral
-
Early exemplar of satellite priors for online mapping.
End-to-End Autonomous Driving
Planning-oriented stacks where perception, prediction, and planning are differentiable and learned jointly.
-
★ Best Paper
The lodestar — perception, prediction, planning fused in one query-based stack.
-
A clear-eyed reset on what the early E2E open-loop numbers really meant.
-
★ Highlight
-
★ E2E AD Challenge — 1st place
-
★ Highlight
Have a paper you think I should read? Send it my way.