Dyna-style model-based reinforcement learning (MBRL) combines model-free agents with predictive transition models through model-based rollouts. This combination raises a critical question: 'When to trust your model?'; i.e., which rollout length results in the model providing useful data? Janner et al. (2019) address this question by gradually increasing rollout lengths throughout the training. While theoretically tempting, uniform model accuracy is a fallacy that collapses at the latest when extrapolating. Instead, we propose asking the question 'Where to trust your model?'. Using inherent model uncertainty to consider local accuracy, we obtain the Model-Based Actor-Critic with Uncertainty-Aware Rollout Adaption (MACURA) algorithm. We propose an easy-to-tune rollout mechanism and demonstrate substantial improvements in data efficiency and performance compared to state-of-the-art deep MBRL methods on the MuJoCo benchmark.
Model-Free Reinforcement Learning (MFRL), leveraging the policy gradient theorem, has demonstrated considerable success in continuous control tasks. However, these approaches are plagued by high gradient variance due to zeroth-order gradient estimation, resulting in suboptimal policies. Conversely, First-Order Model-Based Reinforcement Learning (FO-MBRL) methods employing differentiable simulation provide gradients with reduced variance but are susceptible to sampling error in scenarios involving stiff dynamics, such as physical contact. This paper investigates the source of this error and introduces Adaptive Horizon Actor-Critic (AHAC), an FO-MBRL algorithm that reduces gradient error by adapting the model-based horizon to avoid stiff dynamics. Empirical findings reveal that AHAC outperforms MFRL baselines, attaining 40% more reward across a set of locomotion tasks and efficiently scaling to high-dimensional control environments with improved wall-clock-time efficiency.
Model-based Reinforcement Learning (MBRL) has shown many desirable properties for intelligent control tasks. However, satisfying safety and stability constraints during training and rollout remains an open question. We propose a new Model-based RL framework to enable efficient policy learning with unknown dynamics based on learning model predictive control (LMPC) framework with mathematically provable guarantees of stability. We introduce and explore a novel method for adding safety constraints for model-based RL during training and policy learning. The new stability-augmented framework consists of a neural-network-based learner that learns to construct a Lyapunov function, and a model-based RL agent to consistently complete the tasks while satisfying user-specified constraints given only sub-optimal demonstrations and sparse-cost feedback. We demonstrate the capability of the proposed framework through simulated experiments.
In neuroscience, one of the key behavioral tests for determining whether a subject of study exhibits model-based behavior is to study its adaptiveness to local changes in the environment. In reinforcement learning, however, recent studies have shown that modern model-based agents display poor adaptivity to such changes. The main reason for this is that modern agents are typically designed to improve sample efficiency in single task settings and thus do not take into account the challenges that can arise in other settings. In local adaptation settings, one particularly important challenge is in quickly building and maintaining a sufficiently accurate model after a local change. This is challenging for deep model-based agents as their models and replay buffers are monolithic structures lacking distribution shift handling capabilities. In this study, we show that the conceptually simple idea of partial models can allow deep model-based agents to overcome this challenge and thus allow for building locally adaptive model-based agents. By modeling the different parts of the state space through different models, the agent can not only maintain a model that is accurate across the state space, but it can also quickly adapt it in the presence of a local change in the environment. We demonstrate this by showing that the use of partial models in agents such as deep Dyna-Q, PlaNet and Dreamer can allow for them to effectively adapt to the local changes in their environments.
World models empower model-based agents to interactively explore, reason, and plan within imagined environments for real-world decision-making. However, the high demand for interactivity poses challenges in harnessing recent advancements in video generative models for developing world models at scale. This work introduces Interactive VideoGPT (iVideoGPT), a scalable autoregressive transformer framework that integrates multimodal signals--visual observations, actions, and rewards--into a sequence of tokens, facilitating an interactive experience of agents via next-token prediction. iVideoGPT features a novel compressive tokenization technique that efficiently discretizes high-dimensional visual observations. Leveraging its scalable architecture, we are able to pre-train iVideoGPT on millions of human and robotic manipulation trajectories, establishing a versatile foundation that is adaptable to serve as interactive world models for a wide range of downstream tasks. These include action-conditioned video prediction, visual planning, and model-based reinforcement learning, where iVideoGPT achieves competitive performance compared with state-of-the-art methods. Our work advances the development of interactive general world models, bridging the gap between generative video models and practical model-based reinforcement learning applications. Code and pre-trained models are available at https://thuml.github.io/iVideoGPT.
Achieving energy efficiency in learning is a key challenge for artificial intelligence (AI) computing platforms. Biological systems demonstrate remarkable abilities to learn complex skills quickly and efficiently. Inspired by this, we present a hardware implementation of model-based reinforcement learning (MBRL) using spiking neural networks (SNNs) on mixed-signal analog/digital neuromorphic hardware. This approach leverages the energy efficiency of mixed-signal neuromorphic chips while achieving high sample efficiency through an alternation of online learning, referred to as the "awake" phase, and offline learning, known as the "dreaming" phase. The model proposed includes two symbiotic networks: an agent network that learns by combining real and simulated experiences, and a learned world model network that generates the simulated experiences. We validate the model by training the hardware implementation to play the Atari game Pong. We start from a baseline consisting of an agent network learning without a world model and dreaming, which successfully learns to play the game. By incorporating dreaming, the number of required real game experiences are reduced significantly compared to the baseline. The networks are implemented using a mixed-signal neuromorphic processor, with the readout layers trained using a computer in-the-loop, while the other layers remain fixed. These results pave the way toward energy-efficient neuromorphic learning systems capable of rapid learning in real world applications and use-cases.
In this work we consider Code World Models, world models generated by a Large Language Model (LLM) in the form of Python code for model-based Reinforcement Learning (RL). Calling code instead of LLMs for planning has potential to be more precise, reliable, interpretable, and extremely efficient. However, writing appropriate Code World Models requires the ability to understand complex instructions, to generate exact code with non-trivial logic and to self-debug a long program with feedback from unit tests and environment trajectories. To address these challenges, we propose Generate, Improve and Fix with Monte Carlo Tree Search (GIF-MCTS), a new code generation strategy for LLMs. To test our approach in an offline RL setting, we introduce the Code World Models Benchmark (CWMB), a suite of program synthesis and planning tasks comprised of 18 diverse RL environments paired with corresponding textual descriptions and curated trajectories. GIF-MCTS surpasses all baselines on the CWMB and two other benchmarks, and we show that the Code World Models synthesized with it can be successfully used for planning, resulting in model-based RL agents with greatly improved sample efficiency and inference speed.
Multi-agent reinforcement learning (MARL) algorithms have accomplished remarkable breakthroughs in solving large-scale decision-making tasks. Nonetheless, most existing MARL algorithms are model-free, limiting sample efficiency and hindering their applicability in more challenging scenarios. In contrast, model-based reinforcement learning (MBRL), particularly algorithms integrating planning, such as MuZero, has demonstrated superhuman performance with limited data in many tasks. Hence, we aim to boost the sample efficiency of MARL by adopting model-based approaches. However, incorporating planning and search methods into multi-agent systems poses significant challenges. The expansive action space of multi-agent systems often necessitates leveraging the nearly-independent property of agents to accelerate learning. To tackle this issue, we propose the MAZero algorithm, which combines a centralized model with Monte Carlo Tree Search (MCTS) for policy search. We design a novel network structure to facilitate distributed execution and parameter sharing. To enhance search efficiency in deterministic environments with sizable action spaces, we introduce two novel techniques: Optimistic Search Lambda (OS($\lambda$)) and Advantage-Weighted Policy Optimization (AWPO). Extensive experiments on the SMAC benchmark demonstrate that MAZero outperforms model-free approaches in terms of sample efficiency and provides comparable or better performance than existing model-based methods in terms of both sample and computational efficiency. Our code is available at https://github.com/liuqh16/MAZero.
Visual Model-Based Reinforcement Learning (MBRL) promises to encapsulate agent's knowledge about the underlying dynamics of the environment, enabling learning a world model as a useful planner. However, top MBRL agents such as Dreamer often struggle with visual pixel-based inputs in the presence of exogenous or irrelevant noise in the observation space, due to failure to capture task-specific features while filtering out irrelevant spatio-temporal details. To tackle this problem, we apply a spatio-temporal masking strategy, a bisimulation principle, combined with latent reconstruction, to capture endogenous task-specific aspects of the environment for world models, effectively eliminating non-essential information. Joint training of representations, dynamics, and policy often leads to instabilities. To further address this issue, we develop a Hybrid Recurrent State-Space Model (HRSSM) structure, enhancing state representation robustness for effective policy learning. Our empirical evaluation demonstrates significant performance improvements over existing methods in a range of visually complex control tasks such as Maniskill \cite{gu2023maniskill2} with exogenous distractors from the Matterport environment. Our code is avaliable at https://github.com/bit1029public/HRSSM.
Designing protein nanomaterials of predefined shape and characteristics has the potential to dramatically impact the medical industry. Machine learning (ML) has proven successful in protein design, reducing the need for expensive wet lab experiment rounds. However, challenges persist in efficiently exploring the protein fitness landscapes to identify optimal protein designs. In response, we propose the use of AlphaZero to generate protein backbones, meeting shape and structural scoring requirements. We extend an existing Monte Carlo tree search (MCTS) framework by incorporating a novel threshold-based reward and secondary objectives to improve design precision. This innovation considerably outperforms existing approaches, leading to protein backbones that better respect structural scores. The application of AlphaZero is novel in the context of protein backbone design and demonstrates promising performance. AlphaZero consistently surpasses baseline MCTS by more than 100% in top-down protein design tasks. Additionally, our application of AlphaZero with secondary objectives uncovers further promising outcomes, indicating the potential of model-based reinforcement learning (RL) in navigating the intricate and nuanced aspects of protein design
We present a method that addresses the pain point of long lead-time required to deploy cell-level parameter optimisation policies to new wireless network sites. Given a sequence of action spaces represented by overlapping subsets of cell-level configuration parameters provided by domain experts, we formulate throughput optimisation as Continual Reinforcement Learning of control policies. Simulation results suggest that the proposed system is able to shorten the end-to-end deployment lead-time by two-fold compared to a reinitialise-and-retrain baseline without any drop in optimisation gain.
Effective code optimization in compilers is crucial for computer and software engineering. The success of these optimizations primarily depends on the selection and ordering of the optimization passes applied to the code. While most compilers rely on a fixed sequence of optimization passes, current methods to find the optimal sequence either employ impractically slow search algorithms or learning methods that struggle to generalize to code unseen during training. We introduce CompilerDream, a model-based reinforcement learning approach to general code optimization. CompilerDream comprises a compiler world model that accurately simulates the intrinsic properties of optimization passes and an agent trained on this model to produce effective optimization strategies. By training on a large-scale program dataset, CompilerDream is equipped to serve as a general code optimizer across various application scenarios and source-code languages. Our extensive experiments first highlight CompilerDream's strong optimization capabilities for autotuning, where it leads the CompilerGym leaderboard. More importantly, the zero-shot generalization ability of large-scale trained compiler world model and agent, excels across diverse datasets, surpassing LLVM's built-in optimizations and other state-of-the-art methods in both settings of value prediction and end-to-end code optimization.
This note clarifies some confusions (and perhaps throws out more) around model-based reinforcement learning and their theoretical understanding in the context of deep RL. Main topics of discussion are (1) how to reconcile model-based RL's bad empirical reputation on error compounding with its superior theoretical properties, and (2) the limitations of empirically popular losses. For the latter, concrete counterexamples for the "MuZero loss" are constructed to show that it not only fails in stochastic environments, but also suffers exponential sample complexity in deterministic environments when data provides sufficient coverage.
Model-based reinforcement learning is an effective approach for controlling an unknown system. It is based on a longstanding pipeline familiar to the control community in which one performs experiments on the environment to collect a dataset, uses the resulting dataset to identify a model of the system, and finally performs control synthesis using the identified model. As interacting with the system may be costly and time consuming, targeted exploration is crucial for developing an effective control-oriented model with minimal experimentation. Motivated by this challenge, recent work has begun to study finite sample data requirements and sample efficient algorithms for the problem of optimal exploration in model-based reinforcement learning. However, existing theory and algorithms are limited to model classes which are linear in the parameters. Our work instead focuses on models with nonlinear parameter dependencies, and presents the first finite sample analysis of an active learning algorithm suitable for a general class of nonlinear dynamics. In certain settings, the excess control cost of our algorithm achieves the optimal rate, up to logarithmic factors. We validate our approach in simulation, showcasing the advantage of active, control-oriented exploration for controlling nonlinear systems.
Unsupervised pre-training methods utilizing large and diverse datasets have achieved tremendous success across a range of domains. Recent work has investigated such unsupervised pre-training methods for model-based reinforcement learning (MBRL) but is limited to domain-specific or simulated data. In this paper, we study the problem of pre-training world models with abundant in-the-wild videos for efficient learning of downstream visual control tasks. However, in-the-wild videos are complicated with various contextual factors, such as intricate backgrounds and textured appearance, which precludes a world model from extracting shared world knowledge to generalize better. To tackle this issue, we introduce Contextualized World Models (ContextWM) that explicitly separate context and dynamics modeling to overcome the complexity and diversity of in-the-wild videos and facilitate knowledge transfer between distinct scenes. Specifically, a contextualized extension of the latent dynamics model is elaborately realized by incorporating a context encoder to retain contextual information and empower the image decoder, which encourages the latent dynamics model to concentrate on essential temporal variations. Our experiments show that in-the-wild video pre-training equipped with ContextWM can significantly improve the sample efficiency of MBRL in various domains, including robotic manipulation, locomotion, and autonomous driving. Code is available at this repository: https://github.com/thuml/ContextWM.
We propose a novel model-based reinforcement learning algorithm -- Dynamics Learning and predictive control with Parameterized Actions (DLPA) -- for Parameterized Action Markov Decision Processes (PAMDPs). The agent learns a parameterized-action-conditioned dynamics model and plans with a modified Model Predictive Path Integral control. We theoretically quantify the difference between the generated trajectory and the optimal trajectory during planning in terms of the value they achieved through the lens of Lipschitz Continuity. Our empirical results on several standard benchmarks show that our algorithm achieves superior sample efficiency and asymptotic performance than state-of-the-art PAMDP methods.
Efficiently tackling multiple tasks within complex environment, such as those found in robot manipulation, remains an ongoing challenge in robotics and an opportunity for data-driven solutions, such as reinforcement learning (RL). Model-based RL, by building a dynamic model of the robot, enables data reuse and transfer learning between tasks with the same robot and similar environment. Furthermore, data gathering in robotics is expensive and we must rely on data efficient approaches such as model-based RL, where policy learning is mostly conducted on cheaper simulations based on the learned model. Therefore, the quality of the model is fundamental for the performance of the posterior tasks. In this work, we focus on improving the quality of the model and maintaining the data efficiency by performing active learning of the dynamic model during a preliminary exploration phase based on maximize information gathering. We employ Bayesian neural network models to represent, in a probabilistic way, both the belief and information encoded in the dynamic model during exploration. With our presented strategies we manage to actively estimate the novelty of each transition, using this as the exploration reward. In this work, we compare several Bayesian inference methods for neural networks, some of which have never been used in a robotics context, and evaluate them in a realistic robot manipulation setup. Our experiments show the advantages of our Bayesian model-based RL approach, with similar quality in the results than relevant alternatives with much lower requirements regarding robot execution steps. Unlike related previous studies that focused the validation solely on toy problems, our research takes a step towards more realistic setups, tackling robotic arm end-tasks.
Recent work in reinforcement learning has leveraged symmetries in the model to improve sample efficiency in training a policy. A commonly used simplifying assumption is that the dynamics and reward both exhibit the same symmetry; however, in many real-world environments, the dynamical model exhibits symmetry independent of the reward model. In this paper, we assume only the dynamics exhibit symmetry, extending the scope of problems in reinforcement learning and learning in control theory to which symmetry techniques can be applied. We use Cartan's moving frame method to introduce a technique for learning dynamics that, by construction, exhibit specified symmetries. Numerical experiments demonstrate that the proposed method learns a more accurate dynamical model
Probabilistic world models increase data efficiency of model-based reinforcement learning (MBRL) by guiding the policy with their epistemic uncertainty to improve exploration and acquire new samples. Moreover, the uncertainty-aware learning procedures in probabilistic approaches lead to robust policies that are less sensitive to noisy observations compared to uncertainty unaware solutions. We propose to combine trajectory sampling and deep Gaussian covariance network (DGCN) for a data-efficient solution to MBRL problems in an optimal control setting. We compare trajectory sampling with density-based approximation for uncertainty propagation using three different probabilistic world models; Gaussian processes, Bayesian neural networks, and DGCNs. We provide empirical evidence using four different well-known test environments, that our method improves the sample-efficiency over other combinations of uncertainty propagation methods and probabilistic models. During our tests, we place particular emphasis on the robustness of the learned policies with respect to noisy initial states.
We introduce $\mathcal{L}_1$-MBRL, a control-theoretic augmentation scheme for Model-Based Reinforcement Learning (MBRL) algorithms. Unlike model-free approaches, MBRL algorithms learn a model of the transition function using data and use it to design a control input. Our approach generates a series of approximate control-affine models of the learned transition function according to the proposed switching law. Using the approximate model, control input produced by the underlying MBRL is perturbed by the $\mathcal{L}_1$ adaptive control, which is designed to enhance the robustness of the system against uncertainties. Importantly, this approach is agnostic to the choice of MBRL algorithm, enabling the use of the scheme with various MBRL algorithms. MBRL algorithms with $\mathcal{L}_1$ augmentation exhibit enhanced performance and sample efficiency across multiple MuJoCo environments, outperforming the original MBRL algorithms, both with and without system noise.