ISM: Intelligent Multi-Path Scheduler for Multi-Camera Networked Systems

Immersive applications rely on multi-camera systems streaming multi-view content over wireless links. Modern networks can aggregate paths to sustain high bitrates, yet prior multi-path schedulers remain content-agnostic, failing to exploit the differential frame importance inherent in multi-view transmission. We introduce ISM, a structure-aware reinforcement learning framework for multi-view video delivery over multipath networks. We formulate scheduling as a constrained optimization that jointly minimizes viewport-weighted distortion and synchronization drift under per-path rate limits, then cast it as an MDP. The trained agent decides, per slot, which frames to transmit on which path, conditioned on the video dependency structure and real-time path states. ISM improves multi-view PSNR by 2-3 dB in the low-rate regime and up to 5 dB across the full range, with better inter-view balance. Under non-uniform viewport distributions, viewport-weighted PSNR gains reach 4-5 dB in the low-rate regime.