Our work aims at efficiently leveraging ambiguous demonstrations for the training of a reinforcement learning (RL) agent. An ambiguous demonstration can usually be interpreted in multiple ways, which severely hinders the RL-Agent from learning stably and efficiently. Since an optimal demonstration may also suffer from being ambiguous, previous works that combine RL and learning from demonstration (RLfD works) may not work well. Inspired by how humans handle such situations, we propose to use self-explanation (an agent generates explanations for itself) to recognize valuable high-level relational features as an interpretation of why a successful trajectory is successful. This way, the agent can provide some guidance for its RL learning. Our main contribution is to propose the Self-Explanation for RL from Demonstrations (SERLfD) framework, which can overcome the limitations of traditional RLfD works. Our experimental results show that an RLfD model can be improved by using our SERLfD framework in terms of training stability and performance.

From the earliest years of our lives, humans use language to express our beliefs and desires. Being able to talk to artificial agents about our preferences would thus fulfill a central goal of value alignment. Yet today, we lack computational models explaining such language use. To address this challenge, we formalize learning from language in a contextual bandit setting and ask how a human might communicate preferences over behaviors. We study two distinct types of language: $\textit{instructions}$, which provide information about the desired policy, and $\textit{descriptions}$, which provide information about the reward function. We show that the agent's degree of autonomy determines which form of language is optimal: instructions are better in low-autonomy settings, but descriptions are better when the agent will need to act independently. We then define a pragmatic listener agent that robustly infers the speaker's reward function by reasoning about $\textit{how}$ the speaker expresses themselves. We validate our models with a behavioral experiment, demonstrating that (1) our speaker model predicts human behavior, and (2) our pragmatic listener successfully recovers humans' reward functions. Finally, we show that this form of social learning can integrate with and reduce regret in traditional reinforcement learning. We hope these insights facilitate a shift from developing agents that $\textit{obey}$ language to agents that $\textit{learn}$ from it.

There has been significant recent interest in devising verification techniques for learning-enabled controllers (LECs) that manage safety-critical systems. Given the opacity and lack of interpretability of the neural policies that govern the behavior of such controllers, many existing approaches enforce safety properties through shield, a dynamic monitoring-and-repairing mechanism that ensures a LEC does not emit actions that would violate desired safety conditions. These methods, however, have been shown to have significant scalability limitations because verification costs grow as problem dimensionality and objective complexity increase. In this paper, we propose a new automated verification pipeline capable of synthesizing high-quality safe controllers even when the problem domain involves hundreds of dimensions, or when the desired objective involves stochastic perturbations, liveness considerations, and other complex non-functional properties. Our key insight involves separating safety verification from neural controller training, and using pre-computed verified safety shields to constrain the training process. Experimental results over a range of high-dimensional benchmarks demonstrate the effectiveness of our approach in a range of stochastic linear time-invariant and time-variant systems.

Natural policy gradient methods are popular reinforcement learning methods that improve the stability of policy gradient methods by utilizing second-order approximations to precondition the gradient with the inverse of the Fisher-information matrix. However, to the best of the authors' knowledge, there has not been a study that has investigated the effects of different second-order approximations in a comprehensive and systematic manner. To address this, five different second-order approximations were studied and compared across multiple key metrics including performance, stability, sample efficiency, and computation time. Furthermore, hyperparameters which aren't typically acknowledged in the literature are studied including the effect of different batch sizes and optimizing the critic network with the natural gradient. Experimental results show that on average, improved second-order approximations achieve the best performance and that using properly tuned hyperparameters can lead to large improvements in performance and sample efficiency ranging up to +181%. We also make the code in this study available at https://github.com/gebob19/natural-policy-gradient-reinforcement-learning.

Aerial manipulators (AM) exhibit particularly challenging, non-linear dynamics; the UAV and the manipulator it is carrying form a tightly coupled dynamic system, mutually impacting each other. The mathematical model describing these dynamics forms the core of many solutions in non-linear control and deep reinforcement learning. Traditionally, the formulation of the dynamics involves Euler angle parametrization in the Lagrangian framework or quaternion parametrization in the Newton-Euler framework. The former has the disadvantage of giving birth to singularities and the latter of being algorithmically complex. This work presents a hybrid solution, combining the benefits of both, namely a quaternion approach leveraging the Lagrangian framework, connecting the singularity-free parameterization with the algorithmic simplicity of the Lagrangian approach. We do so by offering detailed insights into the kinematic modeling process and the formulation of the dynamics of a general aerial manipulator. The obtained dynamics model is validated experimentally against a real-time physics engine. A practical application of the obtained dynamics model is shown in the context of a computed torque feedback controller (feedback linearization), where we analyze its real-time capability with increasingly complex models.

Reinforcement learning (RL) algorithms are often categorized as either on-policy or off-policy depending on whether they use data from a target policy of interest or from a different behavior policy. In this paper, we study a subtle distinction between on-policy data and on-policy sampling in the context of the RL sub-problem of policy evaluation. We observe that on-policy sampling may fail to match the expected distribution of on-policy data after observing only a finite number of trajectories and this failure hinders data-efficient policy evaluation. Towards improved data-efficiency, we show how non-i.i.d., off-policy sampling can produce data that more closely matches the expected on-policy data distribution and consequently increases the accuracy of the Monte Carlo estimator for policy evaluation. We introduce a method called Robust On-Policy Sampling and demonstrate theoretically and empirically that it produces data that converges faster to the expected on-policy distribution compared to on-policy sampling. Empirically, we show that this faster convergence leads to lower mean squared error policy value estimates.

With AlphaTensor, DeepMind Technologies has presented an AI system that is supposed to independently find novel, efficient and provably correct algorithms for complex mathematical tasks. AlphaTensor has already identified a new algorithm with which matrix multiplications can be carried out faster than before, as the research team explains in a paper published in the magazine Nature. The team gives the factor as a ten to twenty percent acceleration compared to previous standard methods. AlphaTensor builds on AlphaZero, an AI agent that proved superior to human players in board games such as Chess, Go and Shogi. Founded by British AI researcher, neuroscientist and computer game developer Demosthenes "Demis" Hassabis, the company had already set milestones with AlphaGo and AlphaFold.

Adapting to the changes in transition dynamics is essential in robotic applications. By learning a conditional policy with a compact context, context-aware metareinforcement learning provides a flexible way to adjust behavior according to dynamics changes. However, in real-world applications, the agent may encounter complex dynamics changes. Multiple confounders can influence the transition dynamics, making it challenging to infer accurate context for decision-making. This paper addresses such a challenge by DecOmposed Mutual INformation Optimization (DOMINO) for context learning, which explicitly learns a disentangled context to maximize the mutual information between the context and historical trajectories, while minimizing the state transition prediction error. Our theoretical analysis shows that DOMINO can overcome the underestimation of the mutual information caused by multi-confounded challenges via learning disentangled context and reduce the demand for the number of samples collected in various environments. Extensive experiments show that the context learned by DOMINO benefits both model-based and model-free reinforcement learning algorithms for dynamics generalization in terms of sample efficiency and performance in unseen environments. Open-sourced code is released on our homepage.

Reference-model-assisted deep reinforcement learning (DRL) for inverter-based Volt-Var Control (IB-VVC) in active distribution networks is proposed. We investigate that a large action space increases the learning difficulties of DRL and degrades the optimization performance in the process of generating data and training neural networks. To reduce the action space of DRL, we design a reference-model-assisted DRL approach. We introduce definitions of the reference model, reference-model-based optimization, and reference actions. The reference-model-assisted DRL learns the residual actions between the reference actions and optimal actions, rather than learning the optimal actions directly. Since the residual actions are considerably smaller than the optimal actions for a reference model, we can design a smaller action space for the reference-model-assisted DRL. It reduces the learning difficulties of DRL and optimises the performance of the reference-model-assisted DRL approach. It is noteworthy that the reference-model-assisted DRL approach is compatible with any policy gradient DRL algorithms for continuous action problems. This work takes the soft actor-critic algorithm as an example and designs a reference-model-assisted soft actor-critic algorithm. Simulations show that 1) large action space degrades the performance of DRL in the whole training stage, and 2) reference-model-assisted DRL requires fewer iteration times and returns a better optimization performance.

Offline reinforcement learning (RL) defines the task of learning from a static logged dataset without continually interacting with the environment. The distribution shift between the learned policy and the behavior policy makes it necessary for the value function to stay conservative such that out-of-distribution (OOD) actions will not be severely overestimated. However, existing approaches, penalizing the unseen actions or regularizing with the behavior policy, are too pessimistic, which suppresses the generalization of the value function and hinders the performance improvement. This paper explores mild but enough conservatism for offline learning while not harming generalization. We propose Mildly Conservative Q-learning (MCQ), where OOD actions are actively trained by assigning them proper pseudo Q values. We theoretically show that MCQ induces a policy that behaves at least as well as the behavior policy and no erroneous overestimation will occur for OOD actions. Experimental results on the D4RL benchmarks demonstrate that MCQ achieves remarkable performance compared with prior work. Furthermore, MCQ shows superior generalization ability when transferring from offline to online, and significantly outperforms baselines. Our code is publicly available at https://github.com/dmksjfl/MCQ.