This paper demonstrates a data-driven control approach for demand response in real-life residential buildings. The objective is to optimally schedule the heating cycles of the Domestic Hot Water (DHW) buffer to maximize the self-consumption of the local photovoltaic (PV) production. A model-based reinforcement learning technique is used to tackle the underlying sequential decision-making problem. The proposed algorithm learns the stochastic occupant behavior, predicts the PV production and takes into account the dynamics of the system. A real-life experiment with six residential buildings is performed using this algorithm. The results show that the self-consumption of the PV production is significantly increased, compared to the default thermostat control.
We present a new deep meta reinforcement learner, which we call Deep Episodic Value Iteration (DEVI). DEVI uses a deep neural network to learn a similarity metric for a non-parametric model-based reinforcement learning algorithm. Our model is trained end-to-end via back-propagation. Despite being trained using the model-free Q-learning objective, we show that DEVI's model-based internal structure provides `one-shot' transfer to changes in reward and transition structure, even for tasks with very high-dimensional state spaces.
In recent years, model-free methods that use deep learning have achieved great success in many different reinforcement learning environments. Most successful approaches focus on solving a single task, while multi-task reinforcement learning remains an open problem. In this paper, we present a model based approach to deep reinforcement learning which we use to solve different tasks simultaneously. We show that our approach not only does not degrade but actually benefits from learning multiple tasks. For our model, we also present a new kind of recurrent neural network inspired by residual networks that decouples memory from computation allowing to model complex environments that do not require lots of memory.
Motor adaptation displays a structure-learning effect: adaptation to a new perturbation occurs more quickly when the subject has prior exposure to perturbations with related structure. Although this `learning-to-learn' effect is well documented, its underlying computational mechanisms are poorly understood. We present a new model of motor structure learning, approaching it from the point of view of deep reinforcement learning. Previous work outside of motor control has shown how recurrent neural networks can account for learning-to-learn effects. We leverage this insight to address motor learning, by importing it into the setting of model-based reinforcement learning. We apply the resulting processing architecture to empirical findings from a landmark study of structure learning in target-directed reaching (Braun et al., 2009), and discuss its implications for a wider range of learning-to-learn phenomena.
When an agent cannot represent a perfectly accurate model of its environment's dynamics, model-based reinforcement learning (MBRL) can fail catastrophically. Planning involves composing the predictions of the model; when flawed predictions are composed, even minor errors can compound and render the model useless for planning. Hallucinated Replay (Talvitie 2014) trains the model to "correct" itself when it produces errors, substantially improving MBRL with flawed models. This paper theoretically analyzes this approach, illuminates settings in which it is likely to be effective or ineffective, and presents a novel error bound, showing that a model's ability to self-correct is more tightly related to MBRL performance than one-step prediction error. These results inspire an MBRL algorithm for deterministic MDPs with performance guarantees that are robust to model class limitations.
In this paper we study how to learn stochastic, multimodal transition dynamics in reinforcement learning (RL) tasks. We focus on evaluating transition function estimation, while we defer planning over this model to future work. Stochasticity is a fundamental property of many task environments. However, discriminative function approximators have difficulty estimating multimodal stochasticity. In contrast, deep generative models do capture complex high-dimensional outcome distributions. First we discuss why, amongst such models, conditional variational inference (VI) is theoretically most appealing for model-based RL. Subsequently, we compare different VI models on their ability to learn complex stochasticity on simulated functions, as well as on a typical RL gridworld with multimodal dynamics. Results show VI successfully predicts multimodal outcomes, but also robustly ignores these for deterministic parts of the transition dynamics. In summary, we show a robust method to learn multimodal transitions using function approximation, which is a key preliminary for model-based RL in stochastic domains.
Reinforcement learning is concerned with identifying reward-maximizing behaviour policies in environments that are initially unknown. State-of-the-art reinforcement learning approaches, such as deep Q-networks, are model-free and learn to act effectively across a wide range of environments such as Atari games, but require huge amounts of data. Model-based techniques are more data-efficient, but need to acquire explicit knowledge about the environment. In this paper, we take a step towards using model-based techniques in environments with a high-dimensional visual state space by demonstrating that it is possible to learn system dynamics and the reward structure jointly. Our contribution is to extend a recently developed deep neural network for video frame prediction in Atari games to enable reward prediction as well. To this end, we phrase a joint optimization problem for minimizing both video frame and reward reconstruction loss, and adapt network parameters accordingly. Empirical evaluations on five Atari games demonstrate accurate cumulative reward prediction of up to 200 frames. We consider these results as opening up important directions for model-based reinforcement learning in complex, initially unknown environments.
We designed a grid world task to study human planning and re-planning behavior in an unknown stochastic environment. In our grid world, participants were asked to travel from a random starting point to a random goal position while maximizing their reward. Because they were not familiar with the environment, they needed to learn its characteristics from experience to plan optimally. Later in the task, we randomly blocked the optimal path to investigate whether and how people adjust their original plans to find a detour. To this end, we developed and compared 12 different models. These models were different on how they learned and represented the environment and how they planned to catch the goal. The majority of our participants were able to plan optimally. We also showed that people were capable of revising their plans when an unexpected event occurred. The result from the model comparison showed that the model-based reinforcement learning approach provided the best account for the data and outperformed heuristics in explaining the behavioral data in the re-planning trials.
\begin{abstract} We model individual T2DM patient blood glucose level (BGL) by stochastic process with discrete number of states mainly but not solely governed by medication regimen (e.g. insulin injections). BGL states change otherwise according to various physiological triggers which render a stochastic, statistically unknown, yet assumed to be quasi-stationary, nature of the process. In order to express incentive for being in desired healthy BGL we heuristically define a reward function which returns positive values for desirable BG levels and negative values for undesirable BG levels. The state space consists of sufficient number of states in order to allow for memoryless assumption. This, in turn, allows to formulate Markov Decision Process (MDP), with an objective to maximize the total reward, summarized over a long run. The probability law is found by model-based reinforcement learning (RL) and the optimal insulin treatment policy is retrieved from MDP solution.
We examine the problem of learning mappings from state to state, suitable for use in a model-based reinforcement-learning setting, that simultaneously generalize to novel states and can capture stochastic transitions. We show that currently popular generative adversarial networks struggle to learn these stochastic transition models but a modification to their loss functions results in a powerful learning algorithm for this class of problems.
Reinforcement learning is a powerful paradigm for learning optimal policies from experimental data. However, to find optimal policies, most reinforcement learning algorithms explore all possible actions, which may be harmful for real-world systems. As a consequence, learning algorithms are rarely applied on safety-critical systems in the real world. In this paper, we present a learning algorithm that explicitly considers safety, defined in terms of stability guarantees. Specifically, we extend control-theoretic results on Lyapunov stability verification and show how to use statistical models of the dynamics to obtain high-performance control policies with provable stability certificates. Moreover, under additional regularity assumptions in terms of a Gaussian process prior, we prove that one can effectively and safely collect data in order to learn about the dynamics and thus both improve control performance and expand the safe region of the state space. In our experiments, we show how the resulting algorithm can safely optimize a neural network policy on a simulated inverted pendulum, without the pendulum ever falling down.
Within the context of autonomous driving a model-based reinforcement learning algorithm is proposed for the design of neural network-parameterized controllers. Classical model-based control methods, which include sampling- and lattice-based algorithms and model predictive control, suffer from the trade-off between model complexity and computational burden required for the online solution of expensive optimization or search problems at every short sampling time. To circumvent this trade-off, a 2-step procedure is motivated: first learning of a controller during offline training based on an arbitrarily complicated mathematical system model, before online fast feedforward evaluation of the trained controller. The contribution of this paper is the proposition of a simple gradient-free and model-based algorithm for deep reinforcement learning using task separation with hill climbing (TSHC). In particular, (i) simultaneous training on separate deterministic tasks with the purpose of encoding many motion primitives in a neural network, and (ii) the employment of maximally sparse rewards in combination with virtual velocity constraints (VVCs) in setpoint proximity are advocated.
Model-free deep reinforcement learning algorithms have been shown to be capable of learning a wide range of robotic skills, but typically require a very large number of samples to achieve good performance. Model-based algorithms, in principle, can provide for much more efficient learning, but have proven difficult to extend to expressive, high-capacity models such as deep neural networks. In this work, we demonstrate that medium-sized neural network models can in fact be combined with model predictive control (MPC) to achieve excellent sample complexity in a model-based reinforcement learning algorithm, producing stable and plausible gaits to accomplish various complex locomotion tasks. We also propose using deep neural network dynamics models to initialize a model-free learner, in order to combine the sample efficiency of model-based approaches with the high task-specific performance of model-free methods. We empirically demonstrate on MuJoCo locomotion tasks that our pure model-based approach trained on just random action data can follow arbitrary trajectories with excellent sample efficiency, and that our hybrid algorithm can accelerate model-free learning on high-speed benchmark tasks, achieving sample efficiency gains of 3-5x on swimmer, cheetah, hopper, and ant agents. Videos can be found at https://sites.google.com/view/mbmf
This paper presents a discrete-time option pricing model that is rooted in Reinforcement Learning (RL), and more specifically in the famous Q-Learning method of RL. We construct a risk-adjusted Markov Decision Process for a discrete-time version of the classical Black-Scholes-Merton (BSM) model, where the option price is an optimal Q-function, while the optimal hedge is a second argument of this optimal Q-function, so that both the price and hedge are parts of the same formula. Pricing is done by learning to dynamically optimize risk-adjusted returns for an option replicating portfolio, as in the Markowitz portfolio theory. Using Q-Learning and related methods, once created in a parametric setting, the model is able to go model-free and learn to price and hedge an option directly from data, and without an explicit model of the world. This suggests that RL may provide efficient data-driven and model-free methods for optimal pricing and hedging of options, once we depart from the academic continuous-time limit, and vice versa, option pricing methods developed in Mathematical Finance may be viewed as special cases of model-based Reinforcement Learning. Further, due to simplicity and tractability of our model which only needs basic linear algebra (plus Monte Carlo simulation, if we work with synthetic data), and its close relation to the original BSM model, we suggest that our model could be used for benchmarking of different RL algorithms for financial trading applications
A key challenge in model-based reinforcement learning (RL) is to synthesize computationally efficient and accurate environment models. We show that carefully designed generative models that learn and operate on compact state representations, so-called state-space models, substantially reduce the computational costs for predicting outcomes of sequences of actions. Extensive experiments establish that state-space models accurately capture the dynamics of Atari games from the Arcade Learning Environment from raw pixels. The computational speed-up of state-space models while maintaining high accuracy makes their application in RL feasible: We demonstrate that agents which query these models for decision making outperform strong model-free baselines on the game MSPACMAN, demonstrating the potential of using learned environment models for planning.
We introduce Imagination-Augmented Agents (I2As), a novel architecture for deep reinforcement learning combining model-free and model-based aspects. In contrast to most existing model-based reinforcement learning and planning methods, which prescribe how a model should be used to arrive at a policy, I2As learn to interpret predictions from a learned environment model to construct implicit plans in arbitrary ways, by using the predictions as additional context in deep policy networks. I2As show improved data efficiency, performance, and robustness to model misspecification compared to several baselines.
Model-free reinforcement learning (RL) methods are succeeding in a growing number of tasks, aided by recent advances in deep learning. However, they tend to suffer from high sample complexity, which hinders their use in real-world domains. Alternatively, model-based reinforcement learning promises to reduce sample complexity, but tends to require careful tuning and to date have succeeded mainly in restrictive domains where simple models are sufficient for learning. In this paper, we analyze the behavior of vanilla model-based reinforcement learning methods when deep neural networks are used to learn both the model and the policy, and show that the learned policy tends to exploit regions where insufficient data is available for the model to be learned, causing instability in training. To overcome this issue, we propose to use an ensemble of models to maintain the model uncertainty and regularize the learning process. We further show that the use of likelihood ratio derivatives yields much more stable learning than backpropagation through time. Altogether, our approach Model-Ensemble Trust-Region Policy Optimization (ME-TRPO) significantly reduces the sample complexity compared to model-free deep RL methods on challenging continuous control benchmark tasks.
Research on multi-robot systems has demonstrated promising results in manifold applications and domains. Still, efficiently learning an effective robot behaviors is very difficult, due to unstructured scenarios, high uncertainties, and large state dimensionality (e.g. hyper-redundant and groups of robot). To alleviate this problem, we present Q-CP a cooperative model-based reinforcement learning algorithm, which exploits action values to both (1) guide the exploration of the state space and (2) generate effective policies. Specifically, we exploit Q-learning to attack the curse-of-dimensionality in the iterations of a Monte-Carlo Tree Search. We implement and evaluate Q-CP on different stochastic cooperative (general-sum) games: (1) a simple cooperative navigation problem among 3 robots, (2) a cooperation scenario between a pair of KUKA YouBots performing hand-overs, and (3) a coordination task between two mobile robots entering a door. The obtained results show the effectiveness of Q-CP in the chosen applications, where action values drive the exploration and reduce the computational demand of the planning process while achieving good performance.
We present an algorithm for rapidly learning controllers for robotics systems. The algorithm follows the model-based reinforcement learning paradigm, and improves upon existing algorithms; namely Probabilistic learning in Control (PILCO) and a sample-based version of PILCO with neural network dynamics (Deep-PILCO). We propose training a neural network dynamics model using variational dropout with truncated Log-Normal noise. This allows us to obtain a dynamics model with calibrated uncertainty, which can be used to simulate controller executions via rollouts. We also describe set of techniques, inspired by viewing PILCO as a recurrent neural network model, that are crucial to improve the convergence of the method. We test our method on a variety of benchmark tasks, demonstrating data-efficiency that is competitive with PILCO, while being able to optimize complex neural network controllers. Finally, we assess the performance of the algorithm for learning motor controllers for a six legged autonomous underwater vehicle. This demonstrates the potential of the algorithm for scaling up the dimensionality and dataset sizes, in more complex control tasks.
Existing research studies on vision and language grounding for robot navigation focus on improving model-free deep reinforcement learning (DRL) models in synthetic environments. However, model-free DRL models do not consider the dynamics in the real-world environments, and they often fail to generalize to new scenes. In this paper, we take a radical approach to bridge the gap between synthetic studies and real-world practices---We propose a novel, planned-ahead hybrid reinforcement learning model that combines model-free and model-based reinforcement learning to solve a real-world vision-language navigation task. Our look-ahead module tightly integrates a look-ahead policy model with an environment model that predicts the next state and the reward. Experimental results suggest that our proposed method significantly outperforms the baselines and achieves the best on the real-world Room-to-Room dataset. Moreover, our scalable method is more generalizable when transferring to unseen environments.