CP-SLAM: Collaborative Neural Point-based SLAM System

Authors: Jiarui Hu, Mao Mao, Hujun Bao, Guofeng Zhang, Zhaopeng Cui

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on various datasets demonstrate the superiority of the proposed method in both camera tracking and mapping. ... We evaluate our proposed collaborative SLAM system in two aspects, both single-agent experiments with loop closure and two-agent experiments, of varying sizes and complexity. ... We also conduct ablation studies to show the importance of modules in the proposed system.
Researcher Affiliation Academia Jiarui Hu1, Mao Mao1, Hujun Bao1, Guofeng Zhang1, Zhaopeng Cui1 1State Key Lab of CAD&CG, Zhejiang University
Pseudocode No The paper does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for CP-SLAM is publicly available.
Open Datasets Yes For reconstruction assessment, we utilize the synthetic dataset Replica [35], equipped with a high-quality RGB-D rendering SDK. ... [35] Julian Straub, Thomas Whelan, Lingni Ma, Yufan Chen, Erik Wijmans, Simon Green, Jakob J Engel, Raul Mur-Artal, Carl Ren, Shobhit Verma, et al. The replica dataset: A digital replica of indoor spaces. ar Xiv preprint ar Xiv:1906.05797, 2019.
Dataset Splits No The paper describes the structure of the datasets used (e.g., number of frames, single-agent vs. multi-agent trajectories) but does not provide explicit train/validation/test dataset splits (e.g., percentages or sample counts) for their experimental setup.
Hardware Specification Yes CP-SLAM system runs an RGB-D sequence on an NVIDIA RTX3090 GPU. In the two-agent experiment, we need an additional RTX3090 as the central server.
Software Dependencies No The paper mentions using the "FRNN library" and the "Open3D" library, but it does not specify version numbers for these or any other software components used in the experiments.
Experiment Setup Yes In all our experiments, we set Nnear = 16, Nuni = 4, λ1 = 0.2, Dl = 0.001m, r = 0.15m, ρ = 0.14m, M1 = 3000, M3 = 3136, M2 = 1500. We extract a keyframe every 50 frames and perform map optimization and point cloud supplementation every 10 frames. For single-agent experiments, we optimize the neural field for 200 iterations. For two-agent experiments... we reduce the number of iteration steps to 150.