This is the authors' abstract. We don't add key points for this paper.
In both discrete and continuous domains, model-free reinforcement learning algorithms have been successfully applied to the vast majority of reinforcement learning problems and are the main solution to real world problems. In reinforcement learning problems with continuous action space, state-of-the-art algorithms are extremely sample-inefficient and need lots of training interactions to become proficient, which could be catastrophically expensive and infeasible in real-world problems. So far, frontier algorithms have not used well-defined methods to explore the decision space. Exploring new behaviors is a prerequisite to look for optimal policies. All of the leading algorithms in the field leverage a blind form of exploration added to agent decisions to search for better policies. Such solutions fail to mindfully explore the environment, disrupting the learning process. This makes these algorithms very prone to failing in specific domains. In this research, a novel stochastic Off-Policy Actor-Critic algorithm, TDS for short, is presented. Combining the policy gradient theorem with the deterministic policy gradient, the TDS algorithm can learn how to mindfully explore the environment. The proposed update method enables TDS to learn how to modify the decision stochasticity bonds for each state and action. This is done according to gradients information derived from learning feedbacks. Evaluations in MuJoCo and Box2D tasks show faster convergence or outperform the state-of-the-art algorithms including TD3, SAC, and DDPG in every environment tested.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Asadolahi et al. (2023) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: