RAMS: Resource-Adaptive and Detection-Conditioned Model Switching for Embedded Edge Perception
arXiv:2606.14716v1 Announce Type: cross Abstract: Edge object detection on embedded hardware requires balancing inference latency and detection quality under changing resource pressure. We present RAMS, a lightweight runtime controller that monitors device pressure, calibrates switching thresholds from idle behavior, and dynamically selects among three resident YOLOv8 tiers (NANO/SMALL/MEDIUM at 320/416/640 px) without model-reload latency. RAMS defines five switching policies, including two detection-conditioned variants that prevent aggressive downgrades after recent vulnerable-road-user (VR
The proliferation of edge AI devices and the increasing demand for real-time, resource-efficient perception systems make embedded model switching crucial for practical deployment and safety-critical applications.
This development addresses a fundamental constraint in deploying advanced AI on resource-limited hardware, enabling more robust and reliable autonomous systems at the edge.
The ability to dynamically adapt AI model complexity based on real-time conditions significantly improves performance and safety for embedded edge perception without costly hardware upgrades.
- · Edge AI hardware manufacturers
- · Autonomous vehicle developers
- · Robotics companies
- · Industrial automation
- · Inefficient monolithic AI model developers
- · Companies relying solely on cloud processing for real-time edge use cases
Improved performance and reliability of AI applications on embedded systems, particularly in highly dynamic environments.
Accelerated adoption of AI in compute-constrained edge devices due to enhanced efficiency and safety.
Reduced energy consumption for overall edge AI inference, contributing to lower operational costs and greater sustainability.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI