Recently, robust reinforcement learning (RL) methods designed to handle adversarial input observations have received significant attention, motivated by RL's inherent vulnerabilities. While existing approaches have demonstrated reasonable success, addressing worst-case scenarios over long time horizons requires both minimizing the agent's cumulative rewards for adversaries and training agents to counteract them through alternating learning. However, this process introduces mutual dependencies between the agent and the adversary, making interactions with the environment inefficient and hindering the development of off-policy methods. In this work, we propose a novel off-policy method that eliminates the need for additional environmental interactions by reformulating adversarial learning as a soft-constrained optimization problem. Our approach is theoretically supported by the symmetric property of policy evaluation between the agent and the adversary. The implementation is available at https://github.com/nakanakakosuke/VALT_SAC.

The varying significance of distinct primitive behaviors during the policy learning process has been overlooked by prior model-free RL algorithms. Leveraging this insight, we explore the causal relationship between different action dimensions and rewards to evaluate the significance of various primitive behaviors during training. We introduce a causality-aware entropy term that effectively identifies and prioritizes actions with high potential impacts for efficient exploration. Furthermore, to prevent excessive focus on specific primitive behaviors, we analyze the gradient dormancy phenomenon and introduce a dormancy-guided reset mechanism to further enhance the efficacy of our method. Our proposed algorithm, ACE: Off-policy Actor-critic with Causality-aware Entropy regularization, demonstrates a substantial performance advantage across 29 diverse continuous control tasks spanning 7 domains compared to model-free RL baselines, which underscores the effectiveness, versatility, and efficient sample efficiency of our approach. Benchmark results and videos are available at https://ace-rl.github.io/.

A variety of theoretically-sound policy gradient algorithms exist for the on-policy setting due to the policy gradient theorem, which provides a simplified form for the gradient. The off-policy setting, however, has been less clear due to the existence of multiple objectives and the lack of an explicit off-policy policy gradient theorem. In this work, we unify these objectives into one off-policy objective, and provide a policy gradient theorem for this unified objective. The derivation involves emphatic weightings and interest functions. We show multiple strategies to approximate the gradients, in an algorithm called Actor Critic with Emphatic weightings (ACE). We prove in a counterexample that previous (semi-gradient) off-policy actor-critic methods--particularly Off-Policy Actor-Critic (OffPAC) and Deterministic Policy Gradient (DPG)--converge to the wrong solution whereas ACE finds the optimal solution. We also highlight why these semi-gradient approaches can still perform well in practice, suggesting strategies for variance reduction in ACE. We empirically study several variants of ACE on two classic control environments and an image-based environment designed to illustrate the tradeoffs made by each gradient approximation. We find that by approximating the emphatic weightings directly, ACE performs as well as or better than OffPAC in all settings tested.

We propose a new policy iteration theory as an important extension of soft policy iteration and Soft Actor-Critic (SAC), one of the most efficient model free algorithms for deep reinforcement learning. Supported by the new theory, arbitrary entropy measures that generalize Shannon entropy, such as Tsallis entropy and Renyi entropy, can be utilized to properly randomize action selection while fulfilling the goal of maximizing expected long-term rewards. Our theory gives birth to two new algorithms, i.e., Tsallis entropy Actor-Critic (TAC) and Renyi entropy Actor-Critic (RAC). Theoretical analysis shows that these algorithms can be more effective than SAC. Moreover, they pave the way for us to develop a new Ensemble Actor-Critic (EAC) algorithm in this paper that features the use of a bootstrap mechanism for deep environment exploration as well as a new value-function based mechanism for high-level action selection. Empirically we show that TAC, RAC and EAC can achieve state-of-the-art performance on a range of benchmark control tasks, outperforming SAC and several cutting-edge learning algorithms in terms of both sample efficiency and effectiveness.

Off-policy learning is more unstable compared to on-policy learning in reinforcement learning (RL). One reason for the instability of off-policy learning is a discrepancy between the target ($\pi$) and behavior (b) policy distributions. The discrepancy between $\pi$ and b distributions can be alleviated by employing a smooth variant of the importance sampling (IS), such as the relative importance sampling (RIS). RIS has parameter $\beta\in[0, 1]$ which controls smoothness. To cope with instability, we present the first relative importance sampling-off-policy actor-critic (RIS-Off-PAC) model-free algorithms in RL. In our method, the network yields a target policy (the actor), a value function (the critic) assessing the current policy ($\pi$), and behavior policy. We use action value generated from the behavior policy to train our algorithm rather than from the target policy. We also use deep neural networks to train both actor and critic. We evaluated our algorithm on a number of Open AI Gym benchmark problems and demonstrate better or comparable performance to several state-of-the-art RL baselines.