High Power Laser's (HPL) optimal performance is essential for the success of a wide variety of experimental tasks related to light-matter interactions. Traditionally, HPL parameters are optimised in an automated fashion relying on black-box numerical methods. However, these can be demanding in terms of computational resources and usually disregard transient and complex dynamics. Model-free Deep Reinforcement Learning (DRL) offers a promising alternative framework for optimising HPL performance since it allows to tune the control parameters as a function of system states subject to nonlinear temporal dynamics without requiring an explicit dynamics model of those. Furthermore, DRL aims to find an optimal control policy rather than a static parameter configuration, particularly suitable for dynamic processes involving sequential decision-making. This is particularly relevant as laser systems are typically characterised by dynamic rather than static traits. Hence the need for a strategy to choose the control applied based on the current context instead of one single optimal control configuration. This paper investigates the potential of DRL in improving the efficiency and safety of HPL control systems. We apply this technique to optimise the temporal profile of laser pulses in the L1 pump laser hosted at the ELI Beamlines facility. We show how to adapt DRL to the setting of spectral phase control by solely tuning dispersion coefficients of the spectral phase and reaching pulses similar to transform limited with full-width at half-maximum (FWHM) of ca1.6 ps.

The field of Sequential Decision Making (SDM) provides tools for solving Sequential Decision Processes (SDPs), where an agent must make a series of decisions in order to complete a task or achieve a goal. Historically, two competing SDM paradigms have view for supremacy. Automated Planning (AP) proposes to solve SDPs by performing a reasoning process over a model of the world, often represented symbolically. Conversely, Reinforcement Learning (RL) proposes to learn the solution of the SDP from data, without a world model, and represent the learned knowledge subsymbolically. In the spirit of reconciliation, we provide a review of symbolic, subsymbolic and hybrid methods for SDM. We cover both methods for solving SDPs (e.g., AP, RL and techniques that learn to plan) and for learning aspects of their structure (e.g., world models, state invariants and landmarks). To the best of our knowledge, no other review in the field provides the same scope. As an additional contribution, we discuss what properties an ideal method for SDM should exhibit and argue that neurosymbolic AI is the current approach which most closely resembles this ideal method. Finally, we outline several proposals to advance the field of SDM via the integration of symbolic and subsymbolic AI.

A reliable critic is central to on-policy actor-critic learning. But it becomes challenging to learn a reliable critic in a multi-agent sparse reward scenario due to two factors: 1) The joint action space grows exponentially with the number of agents 2) This, combined with the reward sparseness and environment noise, leads to large sample requirements for accurate learning. We show that regularising the critic with spectral normalization (SN) enables it to learn more robustly, even in multi-agent on-policy sparse reward scenarios. Our experiments show that the regularised critic is quickly able to learn from the sparse rewarding experience in the complex SMAC and RWARE domains. These findings highlight the importance of regularisation in the critic for stable learning.

Deep reinforcement learning algorithms that learn policies by trial-and-error must learn from limited amounts of data collected by actively interacting with the environment. While many prior works have shown that proper regularization techniques are crucial for enabling data-efficient RL, a general understanding of the bottlenecks in data-efficient RL has remained unclear. Consequently, it has been difficult to devise a universal technique that works well across all domains. In this paper, we attempt to understand the primary bottleneck in sample-efficient deep RL by examining several potential hypotheses such as non-stationarity, excessive action distribution shift, and overfitting. We perform thorough empirical analysis on state-based DeepMind control suite (DMC) tasks in a controlled and systematic way to show that high temporal-difference (TD) error on the validation set of transitions is the main culprit that severely affects the performance of deep RL algorithms, and prior methods that lead to good performance do in fact, control the validation TD error to be low. This observation gives us a robust principle for making deep RL efficient: we can hill-climb on the validation TD error by utilizing any form of regularization techniques from supervised learning. We show that a simple online model selection method that targets the validation TD error is effective across state-based DMC and Gym tasks. Reinforcement learning (RL) methods, when combined with high-capacity deep neural net function approximators, have shown promise in domains such as robot manipulation (Andrychowicz et al., 2020), chip placement (Mirhoseini et al., 2020), games (Silver et al., 2016), and data-center cooling (Lazic et al., 2018). Since every unit of active online data collection comes at an expense (e.g., running real robots, chip evaluation using simulation), it is important to develop sample-efficient deep RL algorithms, that can learn efficiently even with limited amount of experience. Devising such efficient RL algorithm has been an important thread of research in recent years (Janner et al., 2019; Chen et al., 2021; Hiraoka et al., 2021). In principle, off-policy RL methods (e.g., SAC (Haarnoja et al., 2018), TD3 (Fujimoto et al., 2018), Rainbow (Hessel et al., 2018)) should provide good sample efficiency, because they make it possible to improve the policy and value functions for many gradient steps per step of data collection. However, this benefit does not appear to be realizable in practice, as taking too many training steps per each collected transition actually harms performance in many environments.

Deep cooperative multi-agent reinforcement learning has demonstrated its remarkable success over a wide spectrum of complex control tasks. However, recent advances in multi-agent learning mainly focus on value decomposition while leaving entity interactions still intertwined, which easily leads to over-fitting on noisy interactions between entities. In this work, we introduce a novel interactiOn Pattern disenTangling (OPT) method, to disentangle not only the joint value function into agent-wise value functions for decentralized execution, but also the entity interactions into interaction prototypes, each of which represents an underlying interaction pattern within a subgroup of the entities. OPT facilitates filtering the noisy interactions between irrelevant entities and thus significantly improves generalizability as well as interpretability. Specifically, OPT introduces a sparse disagreement mechanism to encourage sparsity and diversity among discovered interaction prototypes. Then the model selectively restructures these prototypes into a compact interaction pattern by an aggregator with learnable weights. To alleviate the training instability issue caused by partial observability, we propose to maximize the mutual information between the aggregation weights and the history behaviors of each agent. Experiments on both single-task and multi-task benchmarks demonstrate that the proposed method yields results superior to the state-of-the-art counterparts. Our code is available at https://github.com/liushunyu/OPT.

Real-time control for robotics is a popular research area in the reinforcement learning community. Through the use of techniques such as reward shaping, researchers have managed to train online agents across a multitude of domains. Despite these advances, solving goal-oriented tasks still requires complex architectural changes or hard constraints to be placed on the problem. In this article, we solve the problem of stacking multiple cubes by combining curriculum learning, reward shaping, and a high number of efficiently parallelized environments. We introduce two curriculum learning settings that allow us to separate the complex task into sequential sub-goals, hence enabling the learning of a problem that may otherwise be too difficult. We focus on discussing the challenges encountered while implementing them in a goal-conditioned environment. Finally, we extend the best configuration identified on a higher complexity environment with differently shaped objects.

Rewrite systems [6, 10, 12] have been widely employing equality saturation [9], which is an optimisation methodology that uses a saturated e-graph to represent all possible sequences of rewrite simultaneously, and then extracts the optimal one. As such, optimal results can be achieved by avoiding the phase-ordering problem. However, we observe that when the e-graph is not saturated, it cannot represent all possible rewrite opportunities and therefore the phase-ordering problem is re-introduced during the construction phase of the e-graph. To address this problem, we propose MCTS-GEB, a domain-general rewrite system that applies reinforcement learning (RL) to e-graph construction. At its core, MCTS-GEB uses a Monte Carlo Tree Search (MCTS) [3] to efficiently plan for the optimal e-graph construction, and therefore it can effectively eliminate the phase-ordering problem at the construction phase and achieve better performance within a reasonable time. Evaluation in two different domains shows MCTS-GEB can outperform the state-of-the-art rewrite systems by up to 49x, while the optimisation can generally take less than an hour, indicating MCTS-GEB is a promising building block for the future generation of rewrite systems.

Asymmetrical multiplayer (AMP) game is a popular game genre which involves multiple types of agents competing or collaborating with each other in the game. It is difficult to train powerful agents that can defeat top human players in AMP games by typical self-play training method because of unbalancing characteristics in their asymmetrical environments. We propose asymmetric-evolution training (AET), a novel multi-agent reinforcement learning framework that can train multiple kinds of agents simultaneously in AMP game. We designed adaptive data adjustment (ADA) and environment randomization (ER) to optimize the AET process. We tested our method in a complex AMP game named Tom \& Jerry, and our AIs trained without using any human data can achieve a win rate of 98.5% against top human players over 65 matches. The ablation experiments indicated that the proposed modules are beneficial to the framework.

Dexterous in-hand manipulation for a multi-fingered anthropomorphic hand is extremely difficult because of the high-dimensional state and action spaces, rich contact patterns between the fingers and objects. Even though deep reinforcement learning has made moderate progress and demonstrated its strong potential for manipulation, it is still faced with certain challenges, such as large-scale data collection and high sample complexity. Especially, for some slight change scenes, it always needs to re-collect vast amounts of data and carry out numerous iterations of fine-tuning. Remarkably, humans can quickly transfer learned manipulation skills to different scenarios with little supervision. Inspired by human flexible transfer learning capability, we propose a novel dexterous in-hand manipulation progressive transfer learning framework (PTL) based on efficiently utilizing the collected trajectories and the source-trained dynamics model. This framework adopts progressive neural networks for dynamics model transfer learning on samples selected by a new samples selection method based on dynamics properties, rewards and scores of the trajectories. Experimental results on contact-rich anthropomorphic hand manipulation tasks show that our method can efficiently and effectively learn in-hand manipulation skills with a few online attempts and adjustment learning under the new scene. Compared to learning from scratch, our method can reduce training time costs by 95%.

In this work we propose a coverage planning control approach which allows a mobile agent, equipped with a controllable sensor (i.e., a camera) with limited sensing domain (i.e., finite sensing range and angle of view), to cover the surface area of an object of interest. The proposed approach integrates ray-tracing into the coverage planning process, thus allowing the agent to identify which parts of the scene are visible at any point in time. The problem of integrated ray-tracing and coverage planning control is first formulated as a constrained optimal control problem (OCP), which aims at determining the agent's optimal control inputs over a finite planning horizon, that minimize the coverage time. Efficiently solving the resulting OCP is however very challenging due to non-convex and non-linear visibility constraints. To overcome this limitation, the problem is converted into a Markov decision process (MDP) which is then solved using reinforcement learning. In particular, we show that a controller which follows an optimal control law can be learned using off-policy temporal-difference control (i.e., Q-learning). Extensive numerical experiments demonstrate the effectiveness of the proposed approach for various configurations of the agent and the object of interest.