Reproduce “EdgeNavMamba: Mamba Optimized Object Detection for Energy Efficient Edge Devices” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Romina Aalishah, Mozhgan Navardi, Tinoosh Mohsenin
EdgeNavMamba introduces an energy-efficient deep learning framework that integrates a Mamba-optimized object detector with a reinforcement learning navigation policy, designed for resource-constrained edge devices. The framework achieves significant model compression and up to 73% energy reduction per inference while maintaining robust object detection accuracy and high success rates in simulated goal-directed navigation.
Key insights
- The Mamba architecture, when meticulously optimized with specialized 2D selective scan modules and combined with knowledge distillation, can yield highly efficient vision models suitable for extreme edge device constraints.
- Abstracting visual input into compact bounding box coordinates from a lightweight detector provides a sufficiently rich state representation for an RL agent to learn complex goal-directed navigation tasks, reducing the overall computational burden.
- Empirical energy profiling on physical edge hardware demonstrates that the proposed architectural and compression strategies translate into substantial real-world energy savings (up to 73%) per inference, critical for battery-powered autonomous systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.