Reinforcement Learning(RL) has achieved tremendous development in recent years, but still faces significant obstacles in addressing complex real-life problems due to the issues of poor system generalization, low sample efficiency as well as safety and interpretability concerns. The core reason underlying such dilemmas can be attributed to the fact that most of the work has focused on the computational aspect of value functions or policies using a representational model to describe atomic components of rewards, states and actions etc, thus neglecting the rich high-level declarative domain knowledge of facts, relations and rules that can be either provided a priori or acquired through reasoning over time. Recently, there has been a rapidly growing interest in the use of Knowledge Representation and Reasoning(KRR) methods, usually using logical languages, to enable more abstract representation and efficient learning in RL. In this survey, we provide a preliminary overview on these endeavors that leverage the strengths of KRR to help solving various problems in RL, and discuss the challenging open problems and possible directions for future work in this area.

The emergence of reinforcement learning (RL) methods in traffic signal control tasks has achieved better performance than conventional rule-based approaches. Most RL approaches require the observation of the environment for the agent to decide which action is optimal for a long-term reward. However, in real-world urban scenarios, missing observation of traffic states may frequently occur due to the lack of sensors, which makes existing RL methods inapplicable on road networks with missing observation. In this work, we aim to control the traffic signals in a real-world setting, where some of the intersections in the road network are not installed with sensors and thus with no direct observations around them. To the best of our knowledge, we are the first to use RL methods to tackle the traffic signal control problem in this real-world setting. Specifically, we propose two solutions: the first one imputes the traffic states to enable adaptive control, and the second one imputes both states and rewards to enable adaptive control and the training of RL agents. Through extensive experiments on both synthetic and real-world road network traffic, we reveal that our method outperforms conventional approaches and performs consistently with different missing rates. We also provide further investigations on how missing data influences the performance of our model.

Classical policy search algorithms for robotics typically require performing extensive explorations, which are time-consuming and expensive to implement with real physical platforms. To facilitate the efficient learning of robot manipulation skills, in this work, we propose a new approach comprised of three modules: (1) learning of general prior knowledge with random explorations in simulation, including state representations, dynamic models, and the constrained action space of the task; (2) extraction of a state alignment-based reward function from a single demonstration video; (3) real-time optimization of the imitation policy under systematic safety constraints with sampling-based model predictive control. This solution results in an efficient one-shot imitation-from-video strategy that simplifies the learning and execution of robot skills in real applications. Specifically, we learn priors in a scene of a task family and then deploy the policy in a novel scene immediately following a single demonstration, preventing time-consuming and risky explorations in the environment. As we do not make a strong assumption of dynamic consistency between the scenes, learning priors can be conducted in simulation to avoid collecting data in real-world circumstances. We evaluate the effectiveness of our approach in the context of contact-rich fabric manipulation, which is a common scenario in industrial and domestic tasks. Detailed numerical simulations and real-world hardware experiments reveal that our method can achieve rapid skill acquisition for challenging manipulation tasks.

The Teacher-Student Framework (TSF) is a reinforcement learning setting where a teacher agent guards the training of a student agent by intervening and providing online demonstrations. Assuming optimal, the teacher policy has the perfect timing and capability to intervene in the learning process of the student agent, providing safety guarantee and exploration guidance. Nevertheless, in many real-world settings it is expensive or even impossible to obtain a well-performing teacher policy. In this work, we relax the assumption of a well-performing teacher and develop a new method that can incorporate arbitrary teacher policies with modest or inferior performance. We instantiate an Off-Policy Reinforcement Learning algorithm, termed Teacher-Student Shared Control (TS2C), which incorporates teacher intervention based on trajectory-based value estimation. Theoretical analysis validates that the proposed TS2C algorithm attains efficient exploration and substantial safety guarantee without being affected by the teacher's own performance. Experiments on various continuous control tasks show that our method can exploit teacher policies at different performance levels while maintaining a low training cost. Moreover, the student policy surpasses the imperfect teacher policy in terms of higher accumulated reward in held-out testing environments. Code is available at https://metadriverse.github.io/TS2C.

In deductive domains, three metacognitive knowledge types in ascending order are declarative, procedural, and conditional learning. This work leverages Deep Reinforcement Learning (DRL) in providing adaptive metacognitive interventions to bridge the gap between the three knowledge types and prepare students for future learning across Intelligent Tutoring Systems (ITSs). Students received these interventions that taught how and when to use a backward-chaining (BC) strategy on a logic tutor that supports a default forward-chaining strategy. Six weeks later, we trained students on a probability tutor that only supports BC without interventions. Our results show that on both ITSs, DRL bridged the metacognitive knowledge gap between students and significantly improved their learning performance over their control peers. Furthermore, the DRL policy adapted to the metacognitive development on the logic tutor across declarative, procedural, and conditional students, causing their strategic decisions to be more autonomous.

Deep reinforcement learning (DRL) gives the promise that an agent learns good policy from high-dimensional information, whereas representation learning removes irrelevant and redundant information and retains pertinent information. In this work, we demonstrate that the learned representation of the $Q$-network and its target $Q$-network should, in theory, satisfy a favorable distinguishable representation property. Specifically, there exists an upper bound on the representation similarity of the value functions of two adjacent time steps in a typical DRL setting. However, through illustrative experiments, we show that the learned DRL agent may violate this property and lead to a sub-optimal policy. Therefore, we propose a simple yet effective regularizer called Policy Evaluation with Easy Regularization on Representation (PEER), which aims to maintain the distinguishable representation property via explicit regularization on internal representations. And we provide the convergence rate guarantee of PEER. Implementing PEER requires only one line of code. Our experiments demonstrate that incorporating PEER into DRL can significantly improve performance and sample efficiency. Comprehensive experiments show that PEER achieves state-of-the-art performance on all 4 environments on PyBullet, 9 out of 12 tasks on DMControl, and 19 out of 26 games on Atari. To the best of our knowledge, PEER is the first work to study the inherent representation property of Q-network and its target. Our code is available at https://sites.google.com/view/peer-cvpr2023/.

We study the continuous-time counterpart of Q-learning for reinforcement learning (RL) under the entropy-regularized, exploratory diffusion process formulation introduced by Wang et al. (2020). As the conventional (big) Q-function collapses in continuous time, we consider its first-order approximation and coin the term ``(little) q-function". This function is related to the instantaneous advantage rate function as well as the Hamiltonian. We develop a ``q-learning" theory around the q-function that is independent of time discretization. Given a stochastic policy, we jointly characterize the associated q-function and value function by martingale conditions of certain stochastic processes, in both on-policy and off-policy settings. We then apply the theory to devise different actor-critic algorithms for solving underlying RL problems, depending on whether or not the density function of the Gibbs measure generated from the q-function can be computed explicitly. One of our algorithms interprets the well-known Q-learning algorithm SARSA, and another recovers a policy gradient (PG) based continuous-time algorithm proposed in Jia and Zhou (2022b). Finally, we conduct simulation experiments to compare the performance of our algorithms with those of PG-based algorithms in Jia and Zhou (2022b) and time-discretized conventional Q-learning algorithms.

Reinforcement learning agents naturally learn from extensive exploration. Exploration is costly and can be unsafe in $\textit{safety-critical}$ domains. This paper proposes a novel framework for incorporating domain knowledge to help guide safe exploration and boost sample efficiency. Previous approaches impose constraints, such as regularisation parameters in neural networks, that rely on large sample sets and often are not suitable for safety-critical domains where agents should almost always avoid unsafe actions. In our approach, called $\textit{System III}$, which is inspired by psychologists' notions of the brain's $\textit{System I}$ and $\textit{System II}$, we represent domain expert knowledge of safety in form of first-order logic. We evaluate the satisfaction of these constraints via p-norms in state vector space. In our formulation, constraints are analogous to hazards, objects, and regions of state that have to be avoided during exploration. We evaluated the effectiveness of the proposed method on OpenAI's Gym and Safety-Gym environments. In all tasks, including classic Control and Safety Games, we show that our approach results in safer exploration and sample efficiency.

Artificial intelligence technology has already had a profound impact in various fields such as economy, industry, and education, but still limited. Meta-learning, also known as "learning to learn", provides an opportunity for general artificial intelligence, which can break through the current AI bottleneck. However, meta learning started late and there are fewer projects compare with CV, NLP etc. Each deployment requires a lot of experience to configure the environment, debug code or even rewrite, and the frameworks are isolated. Moreover, there are currently few platforms that focus exclusively on meta-learning, or provide learning materials for novices, for which the threshold is relatively high. Based on this, Awesome-META+, a meta-learning framework integration and learning platform is proposed to solve the above problems and provide a complete and reliable meta-learning framework application and learning platform. The project aims to promote the development of meta-learning and the expansion of the community, including but not limited to the following functions: 1) Complete and reliable meta-learning framework, which can adapt to multi-field tasks such as target detection, image classification, and reinforcement learning. 2) Convenient and simple model deployment scheme which provide convenient meta-learning transfer methods and usage methods to lower the threshold of meta-learning and improve efficiency. 3) Comprehensive researches for learning. 4) Objective and credible performance analysis and thinking.

Abstract--Non-parametric episodic memory can be used to quickly latch onto high-rewarded experience in reinforcement learning tasks. In contrast to parametric deep reinforcement learning approaches in which reward signals need to be backpropagated slowly, these methods only need to discover the solution once, and may then repeatedly solve the task. However, episodic control solutions are stored in discrete tables, and this approach has so far only been applied to discrete action space problems. Deep reinforcement learning (RL) methods have recently Episodic memory is a term that originates from neuroscience demonstrated superhuman performance on a wide range of [9], where it refers to memory that we can quickly tasks, including Gran Turisma [1], StarCraft [2], Go [3], etc. recollect. In the context of RL, this concept has generally been However, in these methods, the weights of neural networks are implemented as a non-parametric (or semi-parametric) table slowly updated over time to match the target predictions based that can be read from and written into rapidly. Information on the encountered reward signal.