Mobile edge computing provides a promising paradigm to satisfy the explosively growing communication and computational demands of power inspection robot

where the collected high-definitional videos are offloaded to the powerful edge servers for data processing and equipment state judgment. However

the lack of global state information（GSI）

capacity-constrained battery

and ultra-reliable and low-latency communications（URLLC） constraint have posed new challenges to task offloading optimization. In this paper

we consider the channel selection problem which is critical to task offloading. With the combined power of reinforcement learning and Lyapunov optimization

we propose a channel selection algorithm combining an energy-awareness

URLLC-awareness

and task-priority-awareness.Without perfect GSI

the algorithm can dynamically optimize channel selection strategies to maximize the utility under the long-term constraints of energy budget and URLLC in a best-effort way. The channel model and electromagnetic interference model derived by measured data in substation are adopted for performance evaluation

and the simulation results verify the effectiveness and reliability of the proposed algorithm.