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..
DynaNav: Dynamic Feature and Layer Selection for Efficient Visual Navigation
Authors: Jiahui Wang, Changhao Chen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments in real-world-based datasets and simulated environments demonstrate the effectiveness of Dyna Nav. Compared to Vi NT, Dyna Nav achieves a 2.26 reduction in FLOPs, 42.3% lower inference time, and 32.8% lower memory usage, while improving navigation performance across four public datasets. |
| Researcher Affiliation | Academia | 1College of Design and Engineering, National University of Singapore 2PEAK-Lab, The Hong Kong University of Science and Technology (Guangzhou) EMAIL EMAIL |
| Pseudocode | No | The paper describes the architecture and processes using figures and text descriptions, but it does not contain a clearly labeled pseudocode block or algorithm. |
| Open Source Code | No | Upon the situation of acceptance, we will consider releasing the code. |
| Open Datasets | Yes | For the benchmark datasets, we select four diverse datasets to assess the performance of our approach under various conditions. These include the Recon dataset [5], which provides medium-speed (2m/s) outdoor data... and the SCAND dataset [70]... Additionally, we include the Go-Stanford dataset [30] and the SACSo N dataset [71]... |
| Dataset Splits | Yes | For each dataset, we randomly split the data into training (80%) and testing (20%) sets. |
| Hardware Specification | No | The paper mentions running simulations and discusses computational costs (FLOPs, inference time, memory usage), but it does not specify any particular GPU or CPU models, memory sizes, or other detailed hardware specifications used for the experiments. |
| Software Dependencies | No | The paper mentions using 'Efficient Net-B0' and 'Adam W' as an optimizer but does not provide specific version numbers for any software, libraries, or frameworks used in the implementation. |
| Experiment Setup | Yes | Table 5: Hyperparameter Settings. This table provides detailed hyperparameters for General (Train Epochs 100, Fine-tuning Epochs 80, Input Resolution 85 64, Training LR 0.0005, Fine-tuning LR 0.0001, Warmup Epochs 3, Optimizer Adam W, LR Scheduler Cosine Annealing, Batch Size 256, λ in loss 0.5, Backbone Type Efficient Net-b0, Hidden Dim 1280), Data (Length of past frames 5, Length of predicted waypoints 5, Max obs-goal distance(meter) 20, Min obs-goal distance(meter) 0), Transformer Decoder (Number of layers 4, Attention Heads 4), Bayesian Optimization (Sim(at, agt t ) Constraint 0.950, Sim(wt, wgt t ) Constraint 0.960, FLOPs Constraint (109) 2.0, Time Constraint (sec) 0.3, Memory Constraint (GB) 14, Optimization Epochs 20, Constraint of Masked Pixels (obs) 2770, Constraint of Masked Pixels (goal) 3400, ξ [0.8,0.5,1.0]), and CARLA Realted (Max Speed 20km/h, Max Distance 900m, Capture Frequency 4Hz). |