The OpenAI Gym project contains hundreds of control problems whose goal is to provide a testbed for reinforcement learning algorithms. One such problem is Freeway-ram-v0, where the observations presented to the agent are 128 bytes of RAM. While the goals of the project are for non-expert AI agents to solve the control problems with general training, in this work, we seek to learn more about the problem, so that we can better evaluate solutions. In particular, we develop on oracle to play the game, so that we may have baselines for success. We present details of the oracle, plus optimal game-playing situations that can be used for training and testing AI agents.
The powerful learning ability of deep neural networks enables reinforcement learning agents to learn competent control policies directly from continuous environments. In theory, to achieve stable performance, neural networks assume i.i.d. inputs, which unfortunately does no hold in the general reinforcement learning paradigm where the training data is temporally correlated and non-stationary. This issue may lead to the phenomenon of "catastrophic interference" and the collapse in performance. In this paper, we present IQ, i.e., interference-aware deep Q-learning, to mitigate catastrophic interference in single-task deep reinforcement learning. Specifically, we resort to online clustering to achieve on-the-fly context division, together with a multi-head network and a knowledge distillation regularization term for preserving the policy of learned contexts. Built upon deep Q networks, IQ consistently boosts the stability and performance when compared to existing methods, verified with extensive experiments on classic control and Atari tasks. The code is publicly available at: https://github.com/Sweety-dm/Interference-aware-Deep-Q-learning.
Discovering successful coordinated behaviors is a central challenge in Multi-Agent Reinforcement Learning (MARL) since it requires exploring a joint action space that grows exponentially with the number of agents. In this paper, we propose a mechanism for achieving sufficient exploration and coordination in a team of agents. Specifically, agents are rewarded for contributing to a more diversified team behavior by employing proper intrinsic motivation functions. To learn meaningful coordination protocols, we structure agents' interactions by introducing a novel framework, where at each timestep, an agent simulates counterfactual rollouts of its policy and, through a sequence of computations, assesses the gap between other agents' current behaviors and their targets. Actions that minimize the gap are considered highly influential and are rewarded. We evaluate our approach on a set of challenging tasks with sparse rewards and partial observability that require learning complex cooperative strategies under a proper exploration scheme, such as the StarCraft Multi-Agent Challenge. Our methods show significantly improved performances over different baselines across all tasks.
In recent years, near-term noisy intermediate scale quantum (NISQ) computing devices have become available. One of the most promising application areas to leverage such NISQ quantum computer prototypes is quantum machine learning. While quantum neural networks are widely studied for supervised learning, quantum reinforcement learning is still just an emerging field of this area. To solve a classical continuous control problem, we use a continuous-variable quantum machine learning approach. We introduce proximal policy optimization for photonic variational quantum agents and also study the effect of the data re-uploading. We present performance assessment via empirical study using Strawberry Fields, a photonic simulator Fock backend and a hybrid training framework connected to an OpenAI Gym environment and TensorFlow. For the restricted CartPole problem, the two variations of the photonic policy learning achieve comparable performance levels and a faster convergence than the baseline classical neural network of same number of trainable parameters.
This paper is an initial endeavor to bridge the gap between powerful Deep Reinforcement Learning methodologies and the problem of exploration/coverage of unknown terrains. Within this scope, MarsExplorer, an openai-gym compatible environment tailored to exploration/coverage of unknown areas, is presented. MarsExplorer translates the original robotics problem into a Reinforcement Learning setup that various off-the-shelf algorithms can tackle. Any learned policy can be straightforwardly applied to a robotic platform without an elaborate simulation model of the robot's dynamics to apply a different learning/adaptation phase. One of its core features is the controllable multi-dimensional procedural generation of terrains, which is the key for producing policies with strong generalization capabilities. Four different state-of-the-art RL algorithms (A3C, PPO, Rainbow, and SAC) are trained on the MarsExplorer environment, and a proper evaluation of their results compared to the average human-level performance is reported. In the follow-up experimental analysis, the effect of the multi-dimensional difficulty setting on the learning capabilities of the best-performing algorithm (PPO) is analyzed. A milestone result is the generation of an exploration policy that follows the Hilbert curve without providing this information to the environment or rewarding directly or indirectly Hilbert-curve-like trajectories. The experimental analysis is concluded by evaluating PPO learned policy algorithm side-by-side with frontier-based exploration strategies. A study on the performance curves revealed that PPO-based policy was capable of performing adaptive-to-the-unknown-terrain sweeping without leaving expensive-to-revisit areas uncovered, underlying the capability of RL-based methodologies to tackle exploration tasks efficiently. The source code can be found at: https://github.com/dimikout3/MarsExplorer.
This paper presents a constrained policy gradient algorithm. We introduce constraints for safe learning with the following steps. First, learning is slowed down (lazy learning) so that the episodic policy change can be computed with the help of the policy gradient theorem and the neural tangent kernel. Then, this enables us the evaluation of the policy at arbitrary states too. In the same spirit, learning can be guided, ensuring safety via augmenting episode batches with states where the desired action probabilities are prescribed. Finally, exogenous discounted sum of future rewards (returns) can be computed at these specific state-action pairs such that the policy network satisfies constraints. Computing the returns is based on solving a system of linear equations (equality constraints) or a constrained quadratic program (inequality constraints, regional constraints). Simulation results suggest that adding constraints (external information) to the learning can improve learning in terms of speed and transparency reasonably if constraints are appropriately selected. The efficiency of the constrained learning was demonstrated with a shallow and wide ReLU network in the Cartpole and Lunar Lander OpenAI gym environments. The main novelty of the paper is giving a practical use of the neural tangent kernel in reinforcement learning.
Previous work on policy learning for Malaria control has often formulated the problem as an optimization problem assuming the objective function and the search space have a specific structure. The problem has been formulated as multi-armed bandits, contextual bandits and a Markov Decision Process in isolation. Furthermore, an emphasis is put on developing new algorithms specific to an instance of Malaria control, while ignoring a plethora of simpler and general algorithms in the literature. In this work, we formally study the formulation of Malaria control and present a comprehensive analysis of several formulations used in the literature. In addition, we implement and analyze several reinforcement learning algorithms in all formulations and compare them to black box optimization. In contrast to previous work, our results show that simple algorithms based on Upper Confidence Bounds are sufficient for learning good Malaria policies, and tend to outperform their more advanced counterparts on the malaria OpenAI Gym environment.
A growing demand is witnessed in both industry and academia for employing Deep Learning (DL) in various domains to solve real-world problems. Deep Reinforcement Learning (DRL) is the application of DL in the domain of Reinforcement Learning (RL). Like any software systems, DRL applications can fail because of faults in their programs. In this paper, we present the first attempt to categorize faults occurring in DRL programs. We manually analyzed 761 artifacts of DRL programs (from Stack Overflow posts and GitHub issues) developed using well-known DRL frameworks (OpenAI Gym, Dopamine, Keras-rl, Tensorforce) and identified faults reported by developers/users. We labeled and taxonomized the identified faults through several rounds of discussions. The resulting taxonomy is validated using an online survey with 19 developers/researchers. To allow for the automatic detection of faults in DRL programs, we have defined a meta-model of DRL programs and developed DRLinter, a model-based fault detection approach that leverages static analysis and graph transformations. The execution flow of DRLinter consists in parsing a DRL program to generate a model conforming to our meta-model and applying detection rules on the model to identify faults occurrences. The effectiveness of DRLinter is evaluated using 15 synthetic DRLprograms in which we injected faults observed in the analyzed artifacts of the taxonomy. The results show that DRLinter can successfully detect faults in all synthetic faulty programs.
This technical report presents panda-gym, a set Reinforcement Learning (RL) environments for the Franka Emika Panda robot integrated with OpenAI Gym. Five tasks are included: reach, push, slide, pick & place and stack. They all follow a Multi-Goal RL framework, allowing to use goal-oriented RL algorithms. To foster open-research, we chose to use the open-source physics engine PyBullet. The implementation chosen for this package allows to define very easily new tasks or new robots. This report also presents a baseline of results obtained with state-of-the-art model-free off-policy algorithms. panda-gym is open-source at https://github.com/qgallouedec/panda-gym.
We present Brax, an open source library for rigid body simulation with a focus on performance and parallelism on accelerators, written in JAX. We present results on a suite of tasks inspired by the existing reinforcement learning literature, but remade in our engine. Additionally, we provide reimplementations of PPO, SAC, ES, and direct policy optimization in JAX that compile alongside our environments, allowing the learning algorithm and the environment processing to occur on the same device, and to scale seamlessly on accelerators. Finally, we include notebooks that facilitate training of performant policies on common OpenAI Gym MuJoCo-like tasks in minutes.
Reinforcement learning is an active research area with a vast number of applications in robotics, and the RoboCup competition is an interesting environment for studying and evaluating reinforcement learning methods. A known difficulty in applying reinforcement learning to robotics is the high number of experience samples required, being the use of simulated environments for training the agents followed by transfer learning to real-world (sim-to-real) a viable path. This article introduces an open-source simulator for the IEEE Very Small Size Soccer and the Small Size League optimized for reinforcement learning experiments. We also propose a framework for creating OpenAI Gym environments with a set of benchmarks tasks for evaluating single-agent and multi-agent robot soccer skills. We then demonstrate the learning capabilities of two state-of-the-art reinforcement learning methods as well as their limitations in certain scenarios introduced in this framework. We believe this will make it easier for more teams to compete in these categories using end-to-end reinforcement learning approaches and further develop this research area.
Episodic memory lets reinforcement learning algorithms remember and exploit promising experience from the past to improve agent performance. Previous works on memory mechanisms show benefits of using episodic-based data structures for discrete action problems in terms of sample-efficiency. The application of episodic memory for continuous control with a large action space is not trivial. Our study aims to answer the question: can episodic memory be used to improve agent's performance in continuous control? Our proposed algorithm combines episodic memory with Actor-Critic architecture by modifying critic's objective. We further improve performance by introducing episodic-based replay buffer prioritization. We evaluate our algorithm on OpenAI gym domains and show greater sample-efficiency compared with the state-of-the art model-free off-policy algorithms.
With the Deep Neural Networks (DNNs) as a powerful function approximator, Deep Reinforcement Learning (DRL) has been excellently demonstrated on robotic control tasks. Compared to DNNs with vanilla artificial neurons, the biologically plausible Spiking Neural Network (SNN) contains a diverse population of spiking neurons, making it naturally powerful on state representation with spatial and temporal information. Based on a hybrid learning framework, where a spike actor-network infers actions from states and a deep critic network evaluates the actor, we propose a Population-coding and Dynamic-neurons improved Spiking Actor Network (PDSAN) for efficient state representation from two different scales: input coding and neuronal coding. For input coding, we apply population coding with dynamically receptive fields to directly encode each input state component. For neuronal coding, we propose different types of dynamic-neurons (containing 1st-order and 2nd-order neuronal dynamics) to describe much more complex neuronal dynamics. Finally, the PDSAN is trained in conjunction with deep critic networks using the Twin Delayed Deep Deterministic policy gradient algorithm (TD3-PDSAN). Extensive experimental results show that our TD3-PDSAN model achieves better performance than state-of-the-art models on four OpenAI gym benchmark tasks. It is an important attempt to improve RL with SNN towards the effective computation satisfying biological plausibility.
The objective of offline RL is to learn optimal policies when a fixed exploratory demonstrations data-set is available and sampling additional observations is impossible (typically if this operation is either costly or rises ethical questions). In order to solve this problem, off the shelf approaches require a properly defined cost function (or its evaluation on the provided data-set), which are seldom available in practice. To circumvent this issue, a reasonable alternative is to query an expert for few optimal demonstrations in addition to the exploratory data-set. The objective is then to learn an optimal policy w.r.t. the expert's latent cost function. Current solutions either solve a behaviour cloning problem (which does not leverage the exploratory data) or a reinforced imitation learning problem (using a fixed cost function that discriminates available exploratory trajectories from expert ones). Inspired by the success of IRL techniques in achieving state of the art imitation performances in online settings, we exploit GAN based data augmentation procedures to construct the first offline IRL algorithm. The obtained policies outperformed the aforementioned solutions on multiple OpenAI gym environments.
Deep reinforcement learning (DRL) agents are trained through trial-and-error interactions with the environment. This leads to a long training time for dense neural networks to achieve good performance. Hence, prohibitive computation and memory resources are consumed. Recently, learning efficient DRL agents has received increasing attention. Yet, current methods focus on accelerating inference time. In this paper, we introduce for the first time a dynamic sparse training approach for deep reinforcement learning to accelerate the training process. The proposed approach trains a sparse neural network from scratch and dynamically adapts its topology to the changing data distribution during training. Experiments on continuous control tasks show that our dynamic sparse agents achieve higher performance than the equivalent dense methods, reduce the parameter count and floating-point operations (FLOPs) by 50%, and have a faster learning speed that enables reaching the performance of dense agents with 40-50% reduction in the training steps.
Biological agents have meaningful interactions with their environment despite the absence of immediate reward signals. In such instances, the agent can learn preferred modes of behaviour that lead to predictable states -- necessary for survival. In this paper, we pursue the notion that this learnt behaviour can be a consequence of reward-free preference learning that ensures an appropriate trade-off between exploration and preference satisfaction. For this, we introduce a model-based Bayesian agent equipped with a preference learning mechanism (pepper) using conjugate priors. These conjugate priors are used to augment the expected free energy planner for learning preferences over states (or outcomes) across time. Importantly, our approach enables the agent to learn preferences that encourage adaptive behaviour at test time. We illustrate this in the OpenAI Gym FrozenLake and the 3D mini-world environments -- with and without volatility. Given a constant environment, these agents learn confident (i.e., precise) preferences and act to satisfy them. Conversely, in a volatile setting, perpetual preference uncertainty maintains exploratory behaviour. Our experiments suggest that learnable (reward-free) preferences entail a trade-off between exploration and preference satisfaction. Pepper offers a straightforward framework suitable for designing adaptive agents when reward functions cannot be predefined as in real environments.
We introduce a framework that abstracts Reinforcement Learning (RL) as a sequence modeling problem. This allows us to draw upon the simplicity and scalability of the Transformer architecture, and associated advances in language modeling such as GPT-x and BERT. In particular, we present Decision Transformer, an architecture that casts the problem of RL as conditional sequence modeling. Unlike prior approaches to RL that fit value functions or compute policy gradients, Decision Transformer simply outputs the optimal actions by leveraging a causally masked Transformer. By conditioning an autoregressive model on the desired return (reward), past states, and actions, our Decision Transformer model can generate future actions that achieve the desired return. Despite its simplicity, Decision Transformer matches or exceeds the performance of state-of-the-art model-free offline RL baselines on Atari, OpenAI Gym, and Key-to-Door tasks.
Human beings, even small children, quickly become adept at figuring out how to use applications on their mobile devices. Learning to use a new app is often achieved via trial-and-error, accelerated by transfer of knowledge from past experiences with like apps. The prospect of building a smarter smartphone - one that can learn how to achieve tasks using mobile apps - is tantalizing. In this paper we explore the use of Reinforcement Learning (RL) with the goal of advancing this aspiration. We introduce an RL-based framework for learning to accomplish tasks in mobile apps. RL agents are provided with states derived from the underlying representation of on-screen elements, and rewards that are based on progress made in the task. Agents can interact with screen elements by tapping or typing. Our experimental results, over a number of mobile apps, show that RL agents can learn to accomplish multi-step tasks, as well as achieve modest generalization across different apps. More generally, we develop a platform which addresses several engineering challenges to enable an effective RL training environment. Our AppBuddy platform is compatible with OpenAI Gym and includes a suite of mobile apps and benchmark tasks that supports a diversity of RL research in the mobile app setting.
Sample-efficient domain adaptation is an open problem in robotics. In this paper, we present affine transport -- a variant of optimal transport, which models the mapping between state transition distributions between the source and target domains with an affine transformation. First, we derive the affine transport framework; then, we extend the basic framework with Procrustes alignment to model arbitrary affine transformations. We evaluate the method in a number of OpenAI Gym sim-to-sim experiments with simulation environments, as well as on a sim-to-real domain adaptation task of a robot hitting a hockeypuck such that it slides and stops at a target position. In each experiment, we evaluate the results when transferring between each pair of dynamics domains. The results show that affine transport can significantly reduce the model adaptation error in comparison to using the original, non-adapted dynamics model.
The gloabal objective of inverse Reinforcement Learning (IRL) is to estimate the unknown cost function of some MDP base on observed trajectories generated by (approximate) optimal policies. The classical approach consists in tuning this cost function so that associated optimal trajectories (that minimise the cumulative discounted cost, i.e. the classical RL loss) are 'similar' to the observed ones. Prior contributions focused on penalising degenerate solutions and improving algorithmic scalability. Quite orthogonally to them, we question the pertinence of characterising optimality with respect to the cumulative discounted cost as it induces an implicit bias against policies with longer mixing times. State of the art value based RL algorithms circumvent this issue by solving for the fixed point of the Bellman optimality operator, a stronger criterion that is not well defined for the inverse problem. To alleviate this bias in IRL, we introduce an alternative training loss that puts more weights on future states which yields a reformulation of the (maximum entropy) IRL problem. The algorithms we devised exhibit enhanced performances (and similar tractability) than off-the-shelf ones in multiple OpenAI gym environments.