The working mechanisms of complex natural systems tend to abide by concise and profound partial differential equations (PDEs). Methods that directly mine equations from data are called PDE discovery, which reveals consistent physical laws and facilitates our adaptive interaction with the natural world. In this paper, an enhanced deep reinforcement-learning framework is proposed to uncover symbolically concise open-form PDEs with little prior knowledge. Particularly, based on a symbol library of basic operators and operands, a structure-aware recurrent neural network agent is designed and seamlessly combined with the sparse regression method to generate concise and open-form PDE expressions. All of the generated PDEs are evaluated by a meticulously designed reward function by balancing fitness to data and parsimony, and updated by the model-based reinforcement learning in an efficient way. Customized constraints and regulations are formulated to guarantee the rationality of PDEs in terms of physics and mathematics. The experiments demonstrate that our framework is capable of mining open-form governing equations of several dynamic systems, even with compound equation terms, fractional structure, and high-order derivatives, with excellent efficiency. Without the need for prior knowledge, this method shows great potential for knowledge discovery in more complicated circumstances with exceptional efficiency and scalability.
Learning representations in the joint domain of vision and touch can improve manipulation dexterity, robustness, and sample-complexity by exploiting mutual information and complementary cues. Here, we present Visuo-Tactile Transformers (VTTs), a novel multimodal representation learning approach suited for model-based reinforcement learning and planning. Our approach extends the Visual Transformer \cite{dosovitskiy2021image} to handle visuo-tactile feedback. Specifically, VTT uses tactile feedback together with self and cross-modal attention to build latent heatmap representations that focus attention on important task features in the visual domain. We demonstrate the efficacy of VTT for representation learning with a comparative evaluation against baselines on four simulated robot tasks and one real world block pushing task. We conduct an ablation study over the components of VTT to highlight the importance of cross-modality in representation learning.
Humans are skillful navigators: We aptly maneuver through new places, realize when we are back at a location we have seen before, and can even conceive of shortcuts that go through parts of our environments we have never visited. Current methods in model-based reinforcement learning on the other hand struggle with generalizing about environment dynamics out of the training distribution. We argue that two principles can help bridge this gap: latent learning and parsimonious dynamics. Humans tend to think about environment dynamics in simple terms -- we reason about trajectories not in reference to what we expect to see along a path, but rather in an abstract latent space, containing information about the places' spatial coordinates. Moreover, we assume that moving around in novel parts of our environment works the same way as in parts we are familiar with. These two principles work together in tandem: it is in the latent space that the dynamics show parsimonious characteristics. We develop a model that learns such parsimonious dynamics. Using a variational objective, our model is trained to reconstruct experienced transitions in a latent space using locally linear transformations, while encouraged to invoke as few distinct transformations as possible. Using our framework, we demonstrate the utility of learning parsimonious latent dynamics models in a range of policy learning and planning tasks.
This paper introduces a simple efficient learning algorithms for general sequential decision making. The algorithm combines Optimism for exploration with Maximum Likelihood Estimation for model estimation, which is thus named OMLE. We prove that OMLE learns the near-optimal policies of an enormously rich class of sequential decision making problems in a polynomial number of samples. This rich class includes not only a majority of known tractable model-based Reinforcement Learning (RL) problems (such as tabular MDPs, factored MDPs, low witness rank problems, tabular weakly-revealing/observable POMDPs and multi-step decodable POMDPs), but also many new challenging RL problems especially in the partially observable setting that were not previously known to be tractable. Notably, the new problems addressed by this paper include (1) observable POMDPs with continuous observation and function approximation, where we achieve the first sample complexity that is completely independent of the size of observation space; (2) well-conditioned low-rank sequential decision making problems (also known as Predictive State Representations (PSRs)), which include and generalize all known tractable POMDP examples under a more intrinsic representation; (3) general sequential decision making problems under SAIL condition, which unifies our existing understandings of model-based RL in both fully observable and partially observable settings. SAIL condition is identified by this paper, which can be viewed as a natural generalization of Bellman/witness rank to address partial observability. This paper also presents a reward-free variant of OMLE algorithm, which learns approximate dynamic models that enable the computation of near-optimal policies for all reward functions simultaneously.
In this paper we present a world model, which learns causal features using the invariance principle. In particular, we use contrastive unsupervised learning to learn the invariant causal features, which enforces invariance across augmentations of irrelevant parts or styles of the observation. The world-model-based reinforcement learning methods independently optimize representation learning and the policy. Thus naive contrastive loss implementation collapses due to a lack of supervisory signals to the representation learning module. We propose an intervention invariant auxiliary task to mitigate this issue. Specifically, we utilize depth prediction to explicitly enforce the invariance and use data augmentation as style intervention on the RGB observation space. Our design leverages unsupervised representation learning to learn the world model with invariant causal features. Our proposed method significantly outperforms current state-of-the-art model-based and model-free reinforcement learning methods on out-of-distribution point navigation tasks on the iGibson dataset. Moreover, our proposed model excels at the sim-to-real transfer of our perception learning module. Finally, we evaluate our approach on the DeepMind control suite and enforce invariance only implicitly since depth is not available. Nevertheless, our proposed model performs on par with the state-of-the-art counterpart.
Commonly in reinforcement learning (RL), rewards are discounted over time using an exponential function to model time preference, thereby bounding the expected long-term reward. In contrast, in economics and psychology, it has been shown that humans often adopt a hyperbolic discounting scheme, which is optimal when a specific task termination time distribution is assumed. In this work, we propose a theory for continuous-time model-based reinforcement learning generalized to arbitrary discount functions. This formulation covers the case in which there is a non-exponential random termination time. We derive a Hamilton-Jacobi-Bellman (HJB) equation characterizing the optimal policy and describe how it can be solved using a collocation method, which uses deep learning for function approximation. Further, we show how the inverse RL problem can be approached, in which one tries to recover properties of the discount function given decision data. We validate the applicability of our proposed approach on two simulated problems. Our approach opens the way for the analysis of human discounting in sequential decision-making tasks.
Propagating state distributions through a generic, uncertain nonlinear dynamical model is known to be intractable and usually begets numerical or analytical approximations. We introduce a method for state prediction, called the Expansion-Compression Unscented Transform, and use it to solve a class of online policy optimization problems. Our proposed algorithm propagates a finite number of sigma points through a state-dependent distribution, which dictates an increase in the number of sigma points at each time step to represent the resulting distribution; this is what we call the expansion operation. To keep the algorithm scalable, we augment the expansion operation with a compression operation based on moment matching, thereby keeping the number of sigma points constant across predictions over multiple time steps. Its performance is empirically shown to be comparable to Monte Carlo but at a much lower computational cost. Under state and control input constraints, the state prediction is subsequently used in tandem with a proposed variant of constrained gradient-descent for online update of policy parameters in a receding horizon fashion. The framework is implemented as a differentiable computational graph for policy training. We showcase our framework for a quadrotor stabilization task as part of a benchmark comparison in safe-control-gym and for optimizing the parameters of a Control Barrier Function based controller in a leader-follower problem.
This paper considers improving wireless communication and computation efficiency in federated learning (FL) via model quantization. In the proposed bitwidth FL scheme, edge devices train and transmit quantized versions of their local FL model parameters to a coordinating server, which aggregates them into a quantized global model and synchronizes the devices. The goal is to jointly determine the bitwidths employed for local FL model quantization and the set of devices participating in FL training at each iteration. We pose this as an optimization problem that aims to minimize the training loss of quantized FL under a per-iteration device sampling budget and delay requirement. However, the formulated problem is difficult to solve without (i) a concrete understanding of how quantization impacts global ML performance and (ii) the ability of the server to construct estimates of this process efficiently. To address the first challenge, we analytically characterize how limited wireless resources and induced quantization errors affect the performance of the proposed FL method. Our results quantify how the improvement of FL training loss between two consecutive iterations depends on the device selection and quantization scheme as well as on several parameters inherent to the model being learned. Then, we show that the FL training process can be described as a Markov decision process and propose a model-based reinforcement learning (RL) method to optimize action selection over iterations. Compared to model-free RL, this model-based RL approach leverages the derived mathematical characterization of the FL training process to discover an effective device selection and quantization scheme without imposing additional device communication overhead. Simulation results show that the proposed FL algorithm can reduce the convergence time.
In this paper, the question how spiking neural network (SNN) learns and fixes in its internal structures a model of external world dynamics is explored. This question is important for implementation of the model-based reinforcement learning (RL), the realistic RL regime where the decisions made by SNN and their evaluation in terms of reward/punishment signals may be separated by significant time interval and sequence of intermediate evaluation-neutral world states. In the present work, I formalize world dynamics as a Markov chain with unknown a priori state transition probabilities, which should be learnt by the network. To make this problem formulation more realistic, I solve it in continuous time, so that duration of every state in the Markov chain may be different and is unknown. It is demonstrated how this task can be accomplished by an SNN with specially designed structure and local synaptic plasticity rules. As an example, we show how this network motif works in the simple but non-trivial world where a ball moves inside a square box and bounces from its walls with a random new direction and velocity.
This paper proposes a novel scoring function for the planning module of MPC-based reinforcement learning methods to address the inherent bias of using the reward function to score trajectories. The proposed method enhances the learning efficiency of existing MPC-based MBRL methods using the discounted sum of values. The method utilizes optimal trajectories to guide policy learning and updates its state-action value function based on real-world and augmented onboard data. The learning efficiency of the proposed method is evaluated in selected MuJoCo Gym environments as well as in learning locomotion skills for a simulated model of the Cassie robot. The results demonstrate that the proposed method outperforms the current state-of-the-art algorithms in terms of learning efficiency and average reward return.
Provably efficient Model-Based Reinforcement Learning (MBRL) based on optimism or posterior sampling (PSRL) is ensured to attain the global optimality asymptotically by introducing the complexity measure of the model. However, the complexity might grow exponentially for the simplest nonlinear models, where global convergence is impossible within finite iterations. When the model suffers a large generalization error, which is quantitatively measured by the model complexity, the uncertainty can be large. The sampled model that current policy is greedily optimized upon will thus be unsettled, resulting in aggressive policy updates and over-exploration. In this work, we propose Conservative Dual Policy Optimization (CDPO) that involves a Referential Update and a Conservative Update. The policy is first optimized under a reference model, which imitates the mechanism of PSRL while offering more stability. A conservative range of randomness is guaranteed by maximizing the expectation of model value. Without harmful sampling procedures, CDPO can still achieve the same regret as PSRL. More importantly, CDPO enjoys monotonic policy improvement and global optimality simultaneously. Empirical results also validate the exploration efficiency of CDPO.
A promising way to improve the sample efficiency of reinforcement learning is model-based methods, in which many explorations and evaluations can happen in the learned models to save real-world samples. However, when the learned model has a non-negligible model error, sequential steps in the model are hard to be accurately evaluated, limiting the model's utilization. This paper proposes to alleviate this issue by introducing multi-step plans to replace multi-step actions for model-based RL. We employ the multi-step plan value estimation, which evaluates the expected discounted return after executing a sequence of action plans at a given state, and updates the policy by directly computing the multi-step policy gradient via plan value estimation. The new model-based reinforcement learning algorithm MPPVE (Model-based Planning Policy Learning with Multi-step Plan Value Estimation) shows a better utilization of the learned model and achieves a better sample efficiency than state-of-the-art model-based RL approaches.
The robustness of any machine learning solution is fundamentally bound by the data it was trained on. One way to generalize beyond the original training is through human-informed augmentation of the original dataset; however, it is impossible to specify all possible failure cases that can occur during deployment. To address this limitation we combine model-based reinforcement learning and model-interpretability methods to propose a solution that self-generates simulated scenarios constrained by environmental concepts and dynamics learned in an unsupervised manner. In particular, an internal model of the agent's environment is conditioned on low-dimensional concept representations of the input space that are sensitive to the agent's actions. We demonstrate this method within a standard realistic driving simulator in a simple point-to-point navigation task, where we show dramatic improvements in one-shot generalization to different instances of specified failure cases as well as zero-shot generalization to similar variations compared to model-based and model-free approaches.
Variational inference (VI) is a specific type of approximate Bayesian inference that approximates an intractable posterior distribution with a tractable one. VI casts the inference problem as an optimization problem, more specifically, the goal is to maximize a lower bound of the logarithm of the marginal likelihood with respect to the parameters of the approximate posterior. Reinforcement learning (RL) on the other hand deals with autonomous agents and how to make them act optimally such as to maximize some notion of expected future cumulative reward. In the non-sequential setting where agents' actions do not have an impact on future states of the environment, RL is covered by contextual bandits and Bayesian optimization. In a proper sequential scenario, however, where agents' actions affect future states, instantaneous rewards need to be carefully traded off against potential long-term rewards. This manuscript shows how the apparently different subjects of VI and RL are linked in two fundamental ways. First, the optimization objective of RL to maximize future cumulative rewards can be recovered via a VI objective under a soft policy constraint in both the non-sequential and the sequential setting. This policy constraint is not just merely artificial but has proven as a useful regularizer in many RL tasks yielding significant improvements in agent performance. And second, in model-based RL where agents aim to learn about the environment they are operating in, the model-learning part can be naturally phrased as an inference problem over the process that governs environment dynamics. We are going to distinguish between two scenarios for the latter: VI when environment states are fully observable by the agent and VI when they are only partially observable through an observation distribution.
We draw on the latest advancements in the physics community to propose a novel method for discovering the governing non-linear dynamics of physical systems in reinforcement learning (RL). We establish that this method is capable of discovering the underlying dynamics using significantly fewer trajectories (as little as one rollout with $\leq 30$ time steps) than state of the art model learning algorithms. Further, the technique learns a model that is accurate enough to induce near-optimal policies given significantly fewer trajectories than those required by model-free algorithms. It brings the benefits of model-based RL without requiring a model to be developed in advance, for systems that have physics-based dynamics. To establish the validity and applicability of this algorithm, we conduct experiments on four classic control tasks. We found that an optimal policy trained on the discovered dynamics of the underlying system can generalize well. Further, the learned policy performs well when deployed on the actual physical system, thus bridging the model to real system gap. We further compare our method to state-of-the-art model-based and model-free approaches, and show that our method requires fewer trajectories sampled on the true physical system compared other methods. Additionally, we explored approximate dynamics models and found that they also can perform well.
Many methods for Model-based Reinforcement learning (MBRL) in Markov decision processes (MDPs) provide guarantees for both the accuracy of the model they can deliver and the learning efficiency. At the same time, state abstraction techniques allow for a reduction of the size of an MDP while maintaining a bounded loss with respect to the original problem. Therefore, it may come as a surprise that no such guarantees are available when combining both techniques, i.e., where MBRL merely observes abstract states. Our theoretical analysis shows that abstraction can introduce a dependence between samples collected online (e.g., in the real world). That means that, without taking this dependence into account, results for MBRL do not directly extend to this setting. Our result shows that we can use concentration inequalities for martingales to overcome this problem. This result makes it possible to extend the guarantees of existing MBRL algorithms to the setting with abstraction. We illustrate this by combining R-MAX, a prototypical MBRL algorithm, with abstraction, thus producing the first performance guarantees for model-based 'RL from Abstracted Observations': model-based reinforcement learning with an abstract model.
Traditional model-based reinforcement learning (RL) methods generate forward rollout traces using the learnt dynamics model to reduce interactions with the real environment. The recent model-based RL method considers the way to learn a backward model that specifies the conditional probability of the previous state given the previous action and the current state to additionally generate backward rollout trajectories. However, in this type of model-based method, the samples derived from backward rollouts and those from forward rollouts are simply aggregated together to optimize the policy via the model-free RL algorithm, which may decrease both the sample efficiency and the convergence rate. This is because such an approach ignores the fact that backward rollout traces are often generated starting from some high-value states and are certainly more instructive for the agent to improve the behavior. In this paper, we propose the backward imitation and forward reinforcement learning (BIFRL) framework where the agent treats backward rollout traces as expert demonstrations for the imitation of excellent behaviors, and then collects forward rollout transitions for policy reinforcement. Consequently, BIFRL empowers the agent to both reach to and explore from high-value states in a more efficient manner, and further reduces the real interactions, making it potentially more suitable for real-robot learning. Moreover, a value-regularized generative adversarial network is introduced to augment the valuable states which are infrequently received by the agent. Theoretically, we provide the condition where BIFRL is superior to the baseline methods. Experimentally, we demonstrate that BIFRL acquires the better sample efficiency and produces the competitive asymptotic performance on various MuJoCo locomotion tasks compared against state-of-the-art model-based methods.
We present an adaptive learning Intelligent Tutoring System, which uses model-based reinforcement learning in the form of contextual bandits to assign learning activities to students. The model is trained on the trajectories of thousands of students in order to maximize their exercise completion rates and continues to learn online, automatically adjusting itself to new activities. A randomized controlled trial with students shows that our model leads to superior completion rates and significantly improved student engagement when compared to other approaches. Our approach is fully-automated unlocking new opportunities for learning experience personalization.
Model-based reinforcement learning (RL) often achieves higher sample efficiency in practice than model-free RL by learning a dynamics model to generate samples for policy learning. Previous works learn a dynamics model that fits under the empirical state-action visitation distribution for all historical policies, i.e., the sample replay buffer. However, in this paper, we observe that fitting the dynamics model under the distribution for \emph{all historical policies} does not necessarily benefit model prediction for the \emph{current policy} since the policy in use is constantly evolving over time. The evolving policy during training will cause state-action visitation distribution shifts. We theoretically analyze how this distribution shift over historical policies affects the model learning and model rollouts. We then propose a novel dynamics model learning method, named \textit{Policy-adapted Dynamics Model Learning (PDML)}. PDML dynamically adjusts the historical policy mixture distribution to ensure the learned model can continually adapt to the state-action visitation distribution of the evolving policy. Experiments on a range of continuous control environments in MuJoCo show that PDML achieves significant improvement in sample efficiency and higher asymptotic performance combined with the state-of-the-art model-based RL methods.
Model-based Reinforcement Learning and Control have demonstrated great potential in various sequential decision making problem domains, including in robotics settings. However, real-world robotics systems often present challenges that limit the applicability of those methods. In particular, we note two problems that jointly happen in many industrial systems: 1) Irregular/asynchronous observations and actions and 2) Dramatic changes in environment dynamics from an episode to another (e.g. varying payload inertial properties). We propose a general framework that overcomes those difficulties by meta-learning adaptive dynamics models for continuous-time prediction and control. The proposed approach is task-agnostic and can be adapted to new tasks in a straight-forward manner. We present evaluations in two different robot simulations and on a real industrial robot.