Perception Utility-Driven Resource Allocation for Multi-Modal Data Transmission

In wireless visual-haptic applications, the coexistence of enhanced mobile broadband (eMBB) video streams and ultra-reliable low-latency communication (URLLC) haptic signals leads to conflicting quality of service (QoS) requirements under limited resources. The uplink carries periodic viewpoint updates and sporadic haptic data, while the downlink delivers high-throughput video and delay-sensitive haptic feedback, creating disparity in traffic patterns. This paper presents a comprehensive analysis of both uplink and downlink transmissions of the video stream and haptic data, particularly the interaction between them. To capture user-perceived quality, we introduce a multi-modal perception utility that jointly models video integrity and end-to-end haptic latency. We then propose a hierarchical proximal policy optimization (HPPO) framework that operates across two timescales: slot-level channel allocation and minislot-level haptic puncturing. Simulation results reveal a clear resource contention between visual and haptic modalities, where prioritized haptic puncturing reduces visual throughput, impacting overall multi-modal perception utility. The proposed HPPO effectively balances this competition, achieving up to 38% higher utility and faster convergence than baseline methods.