Mathematics-Guided Deep Reinforcement Learning for Joint Resource Allocation and Integrated Sensing in Intelligent 6G-Enabled Robotic Systems

Abstract

Sixth-generation (6G) wireless networks are expected to support highly dynamic and heterogeneous applications that require simultaneous improvements in communication efficiency, sensing capability, energy utilization, and computational intelligence. These requirements become particularly challenging in robotic environments, where the mobility of autonomous platforms introduces rapid variations in channel conditions, traffic demands, sensing requirements, and energy consumption. Conventional optimization methods can provide mathematically tractable resource-allocation solutions, but their computational complexity can limit their applicability to rapidly changing network environments. In contrast, deep reinforcement learning (DRL) can learn adaptive resource-allocation policies, but unconstrained learning may produce decisions that violate communication, energy, latency, or sensing requirements.
This paper proposes a mathematics-guided deep reinforcement learning framework for joint resource allocation and integrated sensing in 6G-enabled robotic systems. The proposed framework formulates communication, sensing, energy, interference, and latency requirements as a multi-objective constrained optimization problem. The resulting mathematical formulation is incorporated into the DRL framework through a constraint-aware state representation and reward function. The proposed approach jointly determines communication resources, transmit power, spectrum allocation, reconfigurable intelligent surface configuration, and computational resources while accounting for robotic mobility and sensing requirements. An integrated sensing and communication model is employed to establish the interaction between wireless communication performance and robotic sensing demands. The framework is evaluated through simulation under varying user density, channel conditions, robotic mobility, and network resource availability. Performance is assessed using throughput, latency, energy consumption, spectral efficiency, sensing performance, convergence behaviour, and computational complexity. The proposed framework provides a unified mathematical and learning-based approach for adaptive resource management in intelligent 6G environments and demonstrates the potential of combining constrained optimization with deep reinforcement learning for communication-intensive robotic applications.