Abstract: In this paper, a Markov Decision Process model is built to describe the problem of
task scheduling in cooperative work, and a improved Q-learning algorithm based on Metropolis
rule is present to solve the problem. In the algorithm, Metropolis rule combined with Greedy
Strategy is introduced and a selection in state space is adopted, which accelerate the convergence,
and shorten the running time. Finally, the algorithm is compared to some related algorithms of
other papers, and the algorithm performance is analyzed as well, which indicates the efficiency of
the improved Q-learning algorithm.

Key words: task scheduling, Q-learning, reinforcement learning, simulated annealing