Reinforcement Learning with Success Induced Task Prioritization

Many challenging reinforcement learning (RL) problems require designing a distribution of tasks that can be applied to train effective policies. This specified by the curriculum. A curriculum is meant improve results and accelerate it. We introduce Success Induced Task Prioritization (SITP), framework for automatic learning, where task sequence created based on success rate each task. In this setting, an algorithmically environment instance with unique configuration. The algorithm selects order provide fastest agents. probability selecting any next stage determined evaluating its performance score in previous stages. Experiments were carried out Partially Observable Grid Environment Multiple Agents (POGEMA) Procgen benchmark. demonstrate SITP matches or surpasses other design methods. Our method implemented handful minor modifications standard RL provides useful prioritization minimal computational overhead.