Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
MoE-Gyro: Self-Supervised Over-Range Reconstruction and Denoising for MEMS Gyroscopes
Authors: Feiyang Pan, Shenghe Zheng, Chunyan Yin, Guangbin Dou
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Mo E-Gyro using our proposed ISEBench, demonstrating that our framework significantly extends the measurable range from 450 /s to 1500 /s, reduces Bias Instability by 98.4%, and achieves state-of-the-art performance, effectively addressing the long-standing trade-off in inertial sensing. Our code is available at: https://github.com/2002-Pan/Moe-Gyro |
| Researcher Affiliation | Academia | Feiyang Pan1 , Shenghe Zheng2 , Chunyan Yin1 , Guangbin Dou1 1 Southeast University 2 Harbin Institute of Technology EMAIL EMAIL EMAIL |
| Pseudocode | Yes | Algorithm 1: Mo E route and concat |
| Open Source Code | Yes | Our code is available at: https://github.com/2002-Pan/Moe-Gyro |
| Open Datasets | Yes | Gyro Peak-100 (released with this paper) is a 100 Hz collection captured from the i Phone 14 on-board IMU with ground-truth peak annotations and serves as the sole source for training and evaluating the over-range reconstruction network. For the denoising task we adopt the Visual-Inertial dataset [17] and the Autonomous Platform Inertial dataset [32], both down-sampled to 100 Hz for consistency. |
| Dataset Splits | Yes | We follow an 80 / 20 split of each dataset for training and testing, respectively, and all experiments are executed on a single NVIDIA RTX-4060 GPU. |
| Hardware Specification | Yes | all experiments are executed on a single NVIDIA RTX-4060 GPU. |
| Software Dependencies | No | The experimental details will be provided with the code. |
| Experiment Setup | Yes | The total reconstruction loss is L = L2 + λc Lcorr + λp Lpinn. We fix λp = 0.5 and vary λc from 0.1 to 0.5, then fix λc = 0.5 and vary λp over the same range. As shown in Figure 5(b), the highest PSNR is achieved at (λc, λp) = (0.5, 0.2). Therefore, we adopt λL2 = 1, λc = 0.5, and λp = 0.2 for all over-range reconstruction experiments. |