Neural Radiance Fields (NeRFs) are a powerful tool for implicit scene representations and have been used in various applications, including robotics. However, using NeRFs in sampling-based or Monte-Carlo localisation schemes can be computationally expensive. This research conducts a systematic comparison of different sampling strategies and finds that rendering stable features can significantly reduce the computational cost. Specifically, it can result in a tenfold reduction in the number of network forward passes required, leading to a significant speed improvement.
Publication date: 22 Sep 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.11698
