We address the problem of production planning and distribution in multi-echelon supply chains. We consider uncertain demands and lead times which makes the problem stochastic and non-linear. A Markov Decision Process formulation and a Non-linear Programming model are presented. As a sequential decision-making problem, Deep Reinforcement Learning (RL) is a possible solution approach. This type of technique has gained a lot of attention from Artificial Intelligence and Optimization communities in recent years. Considering the good results obtained with Deep RL approaches in different areas there is a growing interest in applying them in problems from the Operations Research field. We have used a Deep RL technique, namely Proximal Policy Optimization (PPO2), to solve the problem considering uncertain, regular and seasonal demands and constant or stochastic lead times. Experiments are carried out in different scenarios to better assess the suitability of the algorithm. An agent based on a linearized model is used as a baseline. Experimental results indicate that PPO2 is a competitive and adequate tool for this type of problem. PPO2 agent is better than baseline in all scenarios with stochastic lead times (7.3-11.2%), regardless of whether demands are seasonal or not. In scenarios with constant lead times, the PPO2 agent is better when uncertain demands are non-seasonal (2.2-4.7%). The results show that the greater the uncertainty of the scenario, the greater the viability of this type of approach.

In this work we present a preliminary investigation of a novel algorithm called Dyna-T. In reinforcement learning (RL) a planning agent has its own representation of the environment as a model. To discover an optimal policy to interact with the environment, the agent collects experience in a trial and error fashion. Experience can be used for learning a better model or improve directly the value function and policy. Typically separated, Dyna-Q is an hybrid approach which, at each iteration, exploits the real experience to update the model as well as the value function, while planning its action using simulated data from its model. However, the planning process is computationally expensive and strongly depends on the dimensionality of the state-action space. We propose to build a Upper Confidence Tree (UCT) on the simulated experience and search for the best action to be selected during the on-line learning process. We prove the effectiveness of our proposed method on a set of preliminary tests on three testbed environments from Open AI. In contrast to Dyna-Q, Dyna-T outperforms state-of-the-art RL agents in the stochastic environments by choosing a more robust action selection strategy.

Offline reinforcement learning (RL) tasks require the agent to learn from a pre-collected dataset with no further interactions with the environment. Despite the potential to surpass the behavioral policies, RL-based methods are generally impractical due to the training instability and bootstrapping the extrapolation errors, which always require careful hyperparameter tuning via online evaluation. In contrast, offline imitation learning (IL) has no such issues since it learns the policy directly without estimating the value function by bootstrapping. However, IL is usually limited in the capability of the behavioral policy and tends to learn a mediocre behavior from the dataset collected by the mixture of policies. In this paper, we aim to take advantage of IL but mitigate such a drawback. Observing that behavior cloning is able to imitate neighboring policies with less data, we propose \textit{Curriculum Offline Imitation Learning (COIL)}, which utilizes an experience picking strategy for imitating from adaptive neighboring policies with a higher return, and improves the current policy along curriculum stages. On continuous control benchmarks, we compare COIL against both imitation-based and RL-based methods, showing that it not only avoids just learning a mediocre behavior on mixed datasets but is also even competitive with state-of-the-art offline RL methods.

Sub-optimal control policies in intersection traffic signal controllers (TSC) contribute to congestion and lead to negative effects on human health and the environment. Reinforcement learning (RL) for traffic signal control is a promising approach to design better control policies and has attracted considerable research interest in recent years. However, most work done in this area used simplified simulation environments of traffic scenarios to train RL-based TSC. To deploy RL in real-world traffic systems, the gap between simplified simulation environments and real-world applications has to be closed. Therefore, we propose LemgoRL, a benchmark tool to train RL agents as TSC in a realistic simulation environment of Lemgo, a medium-sized town in Germany. In addition to the realistic simulation model, LemgoRL encompasses a traffic signal logic unit that ensures compliance with all regulatory and safety requirements. LemgoRL offers the same interface as the wellknown OpenAI gym toolkit to enable easy deployment in existing research work. To demonstrate the functionality and applicability of LemgoRL, we train a state-of-the-art Deep RL algorithm on a CPU cluster utilizing a framework for distributed and parallel RL and compare its performance with other methods. Our benchmark tool drives the development of RL algorithms towards real-world applications.

Reinforcement learning (RL) methods usually treat reward functions as black boxes. As such, these methods must extensively interact with the environment in order to discover rewards and optimal policies. In most RL applications, however, users have to program the reward function and, hence, there is the opportunity to make the reward function visible – to show the reward function’s code to the RL agent so it can exploit the function’s internal structure to learn optimal policies in a more sample efficient manner. In this paper, we show how to accomplish this idea in two steps. First, we propose reward machines, a type of finite state machine that supports the specification of reward functions while exposing reward function structure. We then describe different methodologies to exploit this structure to support learning, including automated reward shaping, task decomposition, and counterfactual reasoning with off-policy learning. Experiments on tabular and continuous domains, across different tasks and RL agents, show the benefits of exploiting reward structure with respect to sample efficiency and the quality of resultant policies. Finally, by virtue of being a form of finite state machine, reward machines have the expressive power of a regular language and as such support loops, sequences and conditionals, as well as the expression of temporally extended properties typical of linear temporal logic and non-Markovian reward specification.

This paper presents a benchmarking study of some of the state-of-the-art reinforcement learning algorithms used for solving two simulated vision-based robotics problems. The algorithms considered in this study include soft actor-critic (SAC), proximal policy optimization (PPO), interpolated policy gradients (IPG), and their variants with Hindsight Experience replay (HER). The performances of these algorithms are compared against PyBullet's two simulation environments known as KukaDiverseObjectEnv and RacecarZEDGymEnv respectively. The state observations in these environments are available in the form of RGB images and the action space is continuous, making them difficult to solve. A number of strategies are suggested to provide intermediate hindsight goals required for implementing HER algorithm on these problems which are essentially single-goal environments. In addition, a number of feature extraction architectures are proposed to incorporate spatial and temporal attention in the learning process. Through rigorous simulation experiments, the improvement achieved with these components are established. To the best of our knowledge, such a benchmarking study is not available for the above two vision-based robotics problems making it a novel contribution in the field.

Tether-net launched from a chaser spacecraft provides a promising method to capture and dispose of large space debris in orbit. This tether-net system is subject to several sources of uncertainty in sensing and actuation that affect the performance of its net launch and closing control. Earlier reliability-based optimization approaches to design control actions however remain challenging and computationally prohibitive to generalize over varying launch scenarios and target (debris) state relative to the chaser. To search for a general and reliable control policy, this paper presents a reinforcement learning framework that integrates a proximal policy optimization (PPO2) approach with net dynamics simulations. The latter allows evaluating the episodes of net-based target capture, and estimate the capture quality index that serves as the reward feedback to PPO2. Here, the learned policy is designed to model the timing of the net closing action based on the state of the moving net and the target, under any given launch scenario. A stochastic state transition model is considered in order to incorporate synthetic uncertainties in state estimation and launch actuation. Along with notable reward improvement during training, the trained policy demonstrates capture performance (over a wide range of launch/target scenarios) that is close to that obtained with reliability-based optimization run over an individual scenario.

General policy improvement (GPI) and trust-region learning (TRL) are the predominant frameworks within contemporary reinforcement learning (RL), which serve as the core models for solving Markov decision processes (MDPs). Unfortunately, in their mathematical form, they are sensitive to modifications, and thus, the practical instantiations that implement them do not automatically inherit their improvement guarantees. As a result, the spectrum of available rigorous MDP-solvers is narrow. Indeed, many state-of-the-art (SOTA) algorithms, such as TRPO and PPO, are not proven to converge. In this paper, we propose \textsl{mirror learning} -- a general solution to the RL problem. We reveal GPI and TRL to be but small points within this far greater space of algorithms which boasts the monotonic improvement property and converges to the optimal policy. We show that virtually all SOTA algorithms for RL are instances of mirror learning, and thus suggest that their empirical performance is a consequence of their theoretical properties, rather than of approximate analogies. Excitingly, we show that mirror learning opens up a whole new space of policy learning methods with convergence guarantees.

Processing raw sensory inputs is crucial for applying deep RL algorithms to real-world problems. For example, autonomous vehicles must make decisions about how to drive safely given information flowing from cameras, radar, and microphones about the conditions of the road, traffic signals, and other cars and pedestrians. However, direct "end-to-end" RL that maps sensor data to actions (Figure 1, left) can be very difficult because the inputs are high-dimensional, noisy, and contain redundant information. Instead, the challenge is often broken down into two problems (Figure 1, right): (1) extract a representation of the sensory inputs that retains only the relevant information, and (2) perform RL with these representations of the inputs as the system state. Representation learning can extract compact representations of states for RL.

Deep reinforcement learning is an increasingly popular technique for synthesising policies to control an agent's interaction with its environment. There is also growing interest in formally verifying that such policies are correct and execute safely. Progress has been made in this area by building on existing work for verification of deep neural networks and of continuous-state dynamical systems. In this paper, we tackle the problem of verifying probabilistic policies for deep reinforcement learning, which are used to, for example, tackle adversarial environments, break symmetries and manage trade-offs. We propose an abstraction approach, based on interval Markov decision processes, that yields probabilistic guarantees on a policy's execution, and present techniques to build and solve these models using abstract interpretation, mixed-integer linear programming, entropy-based refinement and probabilistic model checking. We implement our approach and illustrate its effectiveness on a selection of reinforcement learning benchmarks.