DynaNav: Dynamic Feature and Layer Selection for Efficient Visual Navigation

2025-09-26

AI Summary

To address the high computational cost and lack of interpretability in Transformer-based decoder models for visual navigation, this paper proposes a Dynamic Feature and Layer Selection (DFLS) framework. The framework jointly integrates trainable hard feature selection, dynamic layer skipping, and early-exit mechanisms, with Bayesian optimization automatically learning optimal exit thresholds to enable scene-adaptive computation path scheduling. Evaluated on four public benchmarks, DFLS reduces FLOPs by 2.26×, inference latency by 42.3%, and memory footprint by 32.8% compared to ViNT, while improving navigation success rate. Its core innovation lies in the synergistic modeling of sparse feature selection and hierarchical dynamic early exiting—thereby simultaneously enhancing computational efficiency, decision interpretability, and end-to-end navigation performance.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Visual navigation is essential for robotics and embodied AI. However, existing foundation models, particularly those with transformer decoders, suffer from high computational overhead and lack interpretability, limiting their deployment in resource-tight scenarios. To address this, we propose DynaNav, a Dynamic Visual Navigation framework that adapts feature and layer selection based on scene complexity. It employs a trainable hard feature selector for sparse operations, enhancing efficiency and interpretability. Additionally, we integrate feature selection into an early-exit mechanism, with Bayesian Optimization determining optimal exit thresholds to reduce computational cost. Extensive experiments in real-world-based datasets and simulated environments demonstrate the effectiveness of DynaNav. Compared to ViNT, DynaNav achieves a 2.26x reduction in FLOPs, 42.3% lower inference time, and 32.8% lower memory usage, while improving navigation performance across four public datasets.

Problem

• Reducing computational overhead in visual navigation models
• Improving model interpretability through dynamic feature selection
• Optimizing resource usage for deployment in constrained scenarios

Innovation

• Dynamic feature and layer selection for efficiency
• Trainable hard feature selector for sparse operations
• Bayesian optimization for early-exit mechanism