TRACE: Ergodic Trajectory Optimization for Active Scene Reconstruction

Researchers introduced TRACE, an ergodic trajectory planning method for active scene reconstruction that moves beyond greedy next-best-view selection.

TRACE builds an online target distribution from uncertainty and visibility, then uses a kernel-ergodic planner with gradient flow and footprint depletion to generate sensor paths.

On the Replica dataset, TRACE improved PSNR by 1.5 dB over the best next-best-view baselines.

The result suggests that globally planned sensing motion can use time more efficiently and produce better 3D reconstructions.