Multi-task Deep Reinforcement Learning: a Combination of Rainbow and DisTraL

Among the previous studies on deep reinforcement learning (DRL) challenges, optimizing data usage efficiency in rich and complex environments has been discussed by many researchers. Recently many scientists have tried to enhance the use of data in reinforcement learning (RL) field. In this regard, data sharing in learning manifesting in transfer learning got high popularity among notable artificial intelligence Procedures. Recent successes in multi-task RL, in which transfer learning plays a major role to use common shared learning data between each separated task effectively, have been the main motivation for this research. On the one hand, DisTraL in multi-task learning has represented potential achievements in application, and on the other hand, by combining Q learning improvements in a hybrid architecture called rainbow, one would be able to benefit from all of the features used. In this study, to achieve higher performance in multi-task DRL scenarios, a rainbow agent in a DisTraL framework has been suggested. Moreover, in order to prove the superiority of the suggested approach over the suggested approach over the standard DisTraL methodology, the results and comparisons have been demonstrated.