Autonomous vehicles have shown promising potential to be a groundbreaking technology for improving the safety of road users. For these vehicles, as well as many other safety-critical robotic technologies, to be deployed in real-world applications, we require algorithms that can generalize well to unseen scenarios and data. Model-based reinforcement learning algorithms (MBRL) have demonstrated state-of-the-art performance and data efficiency across a diverse set of domains. However, these algorithms have also shown susceptibility to changes in the environment and its transition dynamics. In this work, we explore the performance and generalization capabilities of MBRL algorithms for autonomous driving, specifically in the simulated autonomous racing environment, Roboracer (formerly F1Tenth). We frame the head-to-head racing task as a learning problem using contextual Markov decision processes and parameterize the driving behavior of the adversaries using the context of the episode, thereby also parameterizing the transition and reward dynamics. We benchmark the behavior of MBRL algorithms in this environment and propose a novel context-aware extension of the existing literature, cMask. We demonstrate that context-aware MBRL algorithms generalize better to out-of-distribution adversary behaviors relative to context-free approaches. We also demonstrate that cMask displays strong generalization capabilities, as well as further performance improvement relative to other context-aware MBRL approaches when racing against adversaries with in-distribution behaviors.
Learned world models hold significant potential for robotic manipulation, as they can serve as simulator for real-world interactions. While extensive progress has been made in 2D video-based world models, these approaches often lack geometric and spatial reasoning, which is essential for capturing the physical structure of the 3D world. To address this limitation, we introduce iMoWM, a novel interactive world model designed to generate color images, depth maps, and robot arm masks in an autoregressive manner conditioned on actions. To overcome the high computational cost associated with three-dimensional information, we propose MMTokenizer, which unifies multi-modal inputs into a compact token representation. This design enables iMoWM to leverage large-scale pretrained VideoGPT models while maintaining high efficiency and incorporating richer physical information. With its multi-modal representation, iMoWM not only improves the visual quality of future predictions but also serves as an effective simulator for model-based reinforcement learning (MBRL) and facilitates real-world imitation learning. Extensive experiments demonstrate the superiority of iMoWM across these tasks, showcasing the advantages of multi-modal world modeling for robotic manipulation. Homepage: https://xingyoujun.github.io/imowm/
Effective exploration remains a central challenge in model-based reinforcement learning (MBRL), particularly in high-dimensional continuous control tasks where sample efficiency is crucial. A prominent line of recent work leverages learned policies as proposal distributions for Model-Predictive Path Integral (MPPI) planning. Initial approaches update the sampling policy independently of the planner distribution, typically maximizing a learned value function with deterministic policy gradient and entropy regularization. However, because the states encountered during training depend on the MPPI planner, aligning the sampling policy with the planner improves the accuracy of value estimation and long-term performance. To this end, recent methods update the sampling policy by minimizing KL divergence to the planner distribution or by introducing planner-guided regularization into the policy update. In this work, we unify these MPPI-based reinforcement learning methods under a single framework by introducing Policy Optimization-Model Predictive Control (PO-MPC), a family of KL-regularized MBRL methods that integrate the planner's action distribution as a prior in policy optimization. By aligning the learned policy with the planner's behavior, PO-MPC allows more flexibility in the policy updates to trade off Return maximization and KL divergence minimization. We clarify how prior approaches emerge as special cases of this family, and we explore previously unstudied variations. Our experiments show that these extended configurations yield significant performance improvements, advancing the state of the art in MPPI-based RL.
A critical bottleneck in deep reinforcement learning (DRL) is sample inefficiency, as training high-performance agents often demands extensive environmental interactions. Model-based reinforcement learning (MBRL) mitigates this by building world models that simulate environmental dynamics and generate synthetic experience, improving sample efficiency. However, conventional world models process observations holistically, failing to decouple dynamic objects and temporal features from static backgrounds. This approach is computationally inefficient, especially for visual tasks where dynamic objects significantly influence rewards and decision-making performance. To address this, we introduce DyMoDreamer, a novel MBRL algorithm that incorporates a dynamic modulation mechanism to improve the extraction of dynamic features and enrich the temporal information. DyMoDreamer employs differential observations derived from a novel inter-frame differencing mask, explicitly encoding object-level motion cues and temporal dynamics. Dynamic modulation is modeled as stochastic categorical distributions and integrated into a recurrent state-space model (RSSM), enhancing the model's focus on reward-relevant dynamics. Experiments demonstrate that DyMoDreamer sets a new state-of-the-art on the Atari $100$k benchmark with a $156.6$\% mean human-normalized score, establishes a new record of $832$ on the DeepMind Visual Control Suite, and gains a $9.5$\% performance improvement after $1$M steps on the Crafter benchmark. Our code is released at https://github.com/Ultraman-Tiga1/DyMoDreamer.
Model-based reinforcement learning (MBRL) has shown significant potential in robotics due to its high sample efficiency and planning capability. However, extending MBRL to multi-robot cooperation remains challenging due to the complexity of joint dynamics and the reliance on synchronous communication. SeqWM employs independent, autoregressive agent-wise world models to represent joint dynamics, where each agent generates its future trajectory and plans its actions based on the predictions of its predecessors. This design lowers modeling complexity, alleviates the reliance on communication synchronization, and enables the emergence of advanced cooperative behaviors through explicit intention sharing. Experiments in challenging simulated environments (Bi-DexHands and Multi-Quad) demonstrate that SeqWM outperforms existing state-of-the-art model-based and model-free baselines in both overall performance and sample efficiency, while exhibiting advanced cooperative behaviors such as predictive adaptation, temporal alignment, and role division. Furthermore, SeqWM has been success fully deployed on physical quadruped robots, demonstrating its effectiveness in real-world multi-robot systems. Demos and code are available at: https://sites.google.com/view/seqwm-marl
In visuomotor policy learning, the control policy for the robotic agent is derived directly from visual inputs. The typical approach, where a policy and vision encoder are trained jointly from scratch, generalizes poorly to novel visual scene changes. Using pre-trained vision models (PVMs) to inform a policy network improves robustness in model-free reinforcement learning (MFRL). Recent developments in Model-based reinforcement learning (MBRL) suggest that MBRL is more sample-efficient than MFRL. However, counterintuitively, existing work has found PVMs to be ineffective in MBRL. Here, we investigate PVM's effectiveness in MBRL, specifically on generalization under visual domain shifts. We show that, in scenarios with severe shifts, PVMs perform much better than a baseline model trained from scratch. We further investigate the effects of varying levels of fine-tuning of PVMs. Our results show that partial fine-tuning can maintain the highest average task performance under the most extreme distribution shifts. Our results demonstrate that PVMs are highly successful in promoting robustness in visual policy learning, providing compelling evidence for their wider adoption in model-based robotic learning applications.
Traditional RL-based locomotion controllers often suffer from low data efficiency, requiring extensive interaction to achieve robust performance. We present a model-based reinforcement learning (MBRL) framework that improves sample efficiency for quadrupedal locomotion by appending synthetic data to the end of standard rollouts in PPO-based controllers, following the Dyna-Style paradigm. A predictive model, trained alongside the policy, generates short-horizon synthetic transitions that are gradually integrated using a scheduling strategy based on the policy update iterations. Through an ablation study, we identified a strong correlation between sample efficiency and rollout length, which guided the design of our experiments. We validated our approach in simulation on the Unitree Go1 robot and showed that replacing part of the simulated steps with synthetic ones not only mimics extended rollouts but also improves policy return and reduces variance. Finally, we demonstrate that this improvement transfers to the ability to track a wide range of locomotion commands using fewer simulated steps.
There is growing interest in reinforcement learning (RL) methods that leverage the simulator's derivatives to improve learning efficiency. While early gradient-based approaches have demonstrated superior performance compared to derivative-free methods, accessing simulator gradients is often impractical due to their implementation cost or unavailability. Model-based RL (MBRL) can approximate these gradients via learned dynamics models, but the solver efficiency suffers from compounding prediction errors during training rollouts, which can degrade policy performance. We propose an approach that decouples trajectory generation from gradient computation: trajectories are unrolled using a simulator, while gradients are computed via backpropagation through a learned differentiable model of the simulator. This hybrid design enables efficient and consistent first-order policy optimization, even when simulator gradients are unavailable, as well as learning a critic from simulation rollouts, which is more accurate. Our method achieves the sample efficiency and speed of specialized optimizers such as SHAC, while maintaining the generality of standard approaches like PPO and avoiding ill behaviors observed in other first-order MBRL methods. We empirically validate our algorithm on benchmark control tasks and demonstrate its effectiveness on a real Go2 quadruped robot, across both quadrupedal and bipedal locomotion tasks.
The rapid growth of the digital economy and artificial intelligence has transformed cloud data centers into essential infrastructure with substantial energy consumption and carbon emission, necessitating effective energy management. However, existing methods face challenges such as incomplete information, uncertain parameters, and dynamic environments, which hinder their real-world implementation. This paper proposes an adaptive power capping framework tailored to cloud data centers. By dynamically setting the energy consumption upper bound, the power load of data centers can be reshaped to align with the electricity price or other market signals. To this end, we formulate the power capping problem as a partially observable Markov decision process. Subsequently, we develop an uncertainty-aware model-based reinforcement learning (MBRL) method to perceive the cloud data center operational environment and optimize power-capping decisions. By incorporating a two-stage uncertainty-aware optimization algorithm into the MBRL, we improve its adaptability to the ever-changing environment. Additionally, we derive the optimality gap of the proposed scheme under finite iterations, ensuring effective decisions under complex and uncertain scenarios. The numerical experiments validate the effectiveness of the proposed method using a cloud data center operational environment simulator built on real-world production traces from Alibaba, which demonstrates its potential as an efficient energy management solution for cloud data centers.
We introduce Massively Multi-Task Model-Based Policy Optimization (M3PO), a scalable model-based reinforcement learning (MBRL) framework designed to address sample inefficiency in single-task settings and poor generalization in multi-task domains. Existing model-based approaches like DreamerV3 rely on pixel-level generative models that neglect control-centric representations, while model-free methods such as PPO suffer from high sample complexity and weak exploration. M3PO integrates an implicit world model, trained to predict task outcomes without observation reconstruction, with a hybrid exploration strategy that combines model-based planning and model-free uncertainty-driven bonuses. This eliminates the bias-variance trade-off in prior methods by using discrepancies between model-based and model-free value estimates to guide exploration, while maintaining stable policy updates through a trust-region optimizer. M3PO provides an efficient and robust alternative to existing model-based policy optimization approaches and achieves state-of-the-art performance across multiple benchmarks.
Achieving robust robot learning for humanoid locomotion is a fundamental challenge in model-based reinforcement learning (MBRL), where environmental stochasticity and randomness can hinder efficient exploration and learning stability. The environmental, so-called aleatoric, uncertainty can be amplified in high-dimensional action spaces with complex contact dynamics, and further entangled with epistemic uncertainty in the models during learning phases. In this work, we propose DoublyAware, an uncertainty-aware extension of Temporal Difference Model Predictive Control (TD-MPC) that explicitly decomposes uncertainty into two disjoint interpretable components, i.e., planning and policy uncertainties. To handle the planning uncertainty, DoublyAware employs conformal prediction to filter candidate trajectories using quantile-calibrated risk bounds, ensuring statistical consistency and robustness against stochastic dynamics. Meanwhile, policy rollouts are leveraged as structured informative priors to support the learning phase with Group-Relative Policy Constraint (GRPC) optimizers that impose a group-based adaptive trust-region in the latent action space. This principled combination enables the robot agent to prioritize high-confidence, high-reward behavior while maintaining effective, targeted exploration under uncertainty. Evaluated on the HumanoidBench locomotion suite with the Unitree 26-DoF H1-2 humanoid, DoublyAware demonstrates improved sample efficiency, accelerated convergence, and enhanced motion feasibility compared to RL baselines. Our simulation results emphasize the significance of structured uncertainty modeling for data-efficient and reliable decision-making in TD-MPC-based humanoid locomotion learning.
Model-based reinforcement learning (MBRL) has been used to efficiently solve vision-based control tasks in highdimensional image observations. Although recent MBRL algorithms perform well in trained observations, they fail when faced with visual distractions in observations. These task-irrelevant distractions (e.g., clouds, shadows, and light) may be constantly present in real-world scenarios. In this study, we propose a novel self-supervised method, Dream to Generalize (Dr. G), for zero-shot MBRL. Dr. G trains its encoder and world model with dual contrastive learning which efficiently captures task-relevant features among multi-view data augmentations. We also introduce a recurrent state inverse dynamics model that helps the world model to better understand the temporal structure. The proposed methods can enhance the robustness of the world model against visual distractions. To evaluate the generalization performance, we first train Dr. G on simple backgrounds and then test it on complex natural video backgrounds in the DeepMind Control suite, and the randomizing environments in Robosuite. Dr. G yields a performance improvement of 117% and 14% over prior works, respectively. Our code is open-sourced and available at https://github.com/JeongsooHa/DrG.git
This paper addresses the slow policy optimization convergence of Monte Carlo Probabilistic Inference for Learning Control (MC-PILCO), a state-of-the-art model-based reinforcement learning (MBRL) algorithm, by integrating it with iterative Linear Quadratic Regulator (iLQR), a fast trajectory optimization method suitable for nonlinear systems. The proposed method, Exploration-Boosted MC-PILCO (EB-MC-PILCO), leverages iLQR to generate informative, exploratory trajectories and initialize the policy, significantly reducing the number of required optimization steps. Experiments on the cart-pole task demonstrate that EB-MC-PILCO accelerates convergence compared to standard MC-PILCO, achieving up to $\bm{45.9\%}$ reduction in execution time when both methods solve the task in four trials. EB-MC-PILCO also maintains a $\bm{100\%}$ success rate across trials while solving the task faster, even in cases where MC-PILCO converges in fewer iterations.
The Cross-Entropy Method (CEM) is a widely adopted trajectory optimizer in model-based reinforcement learning (MBRL), but its unimodal sampling strategy often leads to premature convergence in multimodal landscapes. In this work, we propose Bregman Centroid Guided CEM ($\mathcal{BC}$-EvoCEM), a lightweight enhancement to ensemble CEM that leverages $\textit{Bregman centroids}$ for principled information aggregation and diversity control. $\textbf{$\mathcal{BC}$-EvoCEM}$ computes a performance-weighted Bregman centroid across CEM workers and updates the least contributing ones by sampling within a trust region around the centroid. Leveraging the duality between Bregman divergences and exponential family distributions, we show that $\textbf{$\mathcal{BC}$-EvoCEM}$ integrates seamlessly into standard CEM pipelines with negligible overhead. Empirical results on synthetic benchmarks, a cluttered navigation task, and full MBRL pipelines demonstrate that $\textbf{$\mathcal{BC}$-EvoCEM}$ enhances both convergence and solution quality, providing a simple yet effective upgrade for CEM.
Recent advances in model-based reinforcement learning (MBRL) have achieved super-human level performance on the Atari100k benchmark, driven by reinforcement learning agents trained on powerful diffusion world models. However, we identify that the current aggregates mask a major performance asymmetry: MBRL agents dramatically outperform humans in some tasks despite drastically underperforming in others, with the former inflating the aggregate metrics. This is especially pronounced in pixel-based agents trained with diffusion world models. In this work, we address the pronounced asymmetry observed in pixel-based agents as an initial attempt to reverse the worrying upward trend observed in them. We address the problematic aggregates by delineating all tasks as Agent-Optimal or Human-Optimal and advocate for equal importance on metrics from both sets. Next, we hypothesize this pronounced asymmetry is due to the lack of temporally-structured latent space trained with the World Model objective in pixel-based methods. Lastly, to address this issue, we propose Joint Embedding DIffusion (JEDI), a novel latent diffusion world model trained end-to-end with the self-consistency objective. JEDI outperforms SOTA models in human-optimal tasks while staying competitive across the Atari100k benchmark, and runs 3 times faster with 43% lower memory than the latest pixel-based diffusion baseline. Overall, our work rethinks what it truly means to cross human-level performance in Atari100k.
Model-based reinforcement learning (MBRL) offers an intuitive way to increase the sample efficiency of model-free RL methods by simultaneously training a world model that learns to predict the future. MBRL methods have progressed by largely prioritising the actor; optimising the world model learning has been neglected meanwhile. Improving the fidelity of the world model and reducing its time to convergence can yield significant downstream benefits, one of which is improving the ensuing performance of any actor it may train. We propose a novel approach that anticipates and actively seeks out high-entropy states using short-horizon latent predictions generated by the world model, offering a principled alternative to traditional curiosity-driven methods that chase once-novel states well after they were stumbled into. While many model predictive control (MPC) based methods offer similar alternatives, they typically lack commitment, synthesising multi step plans after every step. To mitigate this, we present a hierarchical planner that dynamically decides when to replan, planning horizon length, and the weighting between reward and entropy. While our method can theoretically be applied to any model that trains its own actors with solely model generated data, we have applied it to just Dreamer as a proof of concept. Our method finishes the Miniworld procedurally generated mazes 50% faster than base Dreamer at convergence and the policy trained in imagination converges in only 60% of the environment steps that base Dreamer needs.
Reinforcement Learning (RL) can mitigate the causal confusion and distribution shift inherent to imitation learning (IL). However, applying RL to end-to-end autonomous driving (E2E-AD) remains an open problem for its training difficulty, and IL is still the mainstream paradigm in both academia and industry. Recently Model-based Reinforcement Learning (MBRL) have demonstrated promising results in neural planning; however, these methods typically require privileged information as input rather than raw sensor data. We fill this gap by designing Raw2Drive, a dual-stream MBRL approach. Initially, we efficiently train an auxiliary privileged world model paired with a neural planner that uses privileged information as input. Subsequently, we introduce a raw sensor world model trained via our proposed Guidance Mechanism, which ensures consistency between the raw sensor world model and the privileged world model during rollouts. Finally, the raw sensor world model combines the prior knowledge embedded in the heads of the privileged world model to effectively guide the training of the raw sensor policy. Raw2Drive is so far the only RL based end-to-end method on CARLA Leaderboard 2.0, and Bench2Drive and it achieves state-of-the-art performance.
Continuous time systems are often modeled using discrete time dynamics but this requires a small simulation step to maintain accuracy. In turn, this requires a large planning horizon which leads to computationally demanding planning problems and reduced performance. Previous work in model free reinforcement learning has partially addressed this issue using action repeats where a policy is learned to determine a discrete action duration. Instead we propose to control the continuous decision timescale directly by using temporally-extended actions and letting the planner treat the duration of the action as an additional optimization variable along with the standard action variables. This additional structure has multiple advantages. It speeds up simulation time of trajectories and, importantly, it allows for deep horizon search in terms of primitive actions while using a shallow search depth in the planner. In addition, in the model based reinforcement learning (MBRL) setting, it reduces compounding errors from model learning and improves training time for models. We show that this idea is effective and that the range for action durations can be automatically selected using a multi-armed bandit formulation and integrated into the MBRL framework. An extensive experimental evaluation both in planning and in MBRL, shows that our approach yields faster planning, better solutions, and that it enables solutions to problems that are not solved in the standard formulation.
Offline reinforcement learning (RL) offers a powerful paradigm for data-driven control. Compared to model-free approaches, offline model-based RL (MBRL) explicitly learns a world model from a static dataset and uses it as a surrogate simulator, improving data efficiency and enabling potential generalization beyond the dataset support. However, most existing offline MBRL methods follow a two-stage training procedure: first learning a world model by maximizing the likelihood of the observed transitions, then optimizing a policy to maximize its expected return under the learned model. This objective mismatch results in a world model that is not necessarily optimized for effective policy learning. Moreover, we observe that policies learned via offline MBRL often lack robustness during deployment, and small adversarial noise in the environment can lead to significant performance degradation. To address these, we propose a framework that dynamically adapts the world model alongside the policy under a unified learning objective aimed at improving robustness. At the core of our method is a maximin optimization problem, which we solve by innovatively utilizing Stackelberg learning dynamics. We provide theoretical analysis to support our design and introduce computationally efficient implementations. We benchmark our algorithm on twelve noisy D4RL MuJoCo tasks and three stochastic Tokamak Control tasks, demonstrating its state-of-the-art performance.
The goal of offline reinforcement learning (RL) is to extract a high-performance policy from the fixed datasets, minimizing performance degradation due to out-of-distribution (OOD) samples. Offline model-based RL (MBRL) is a promising approach that ameliorates OOD issues by enriching state-action transitions with augmentations synthesized via a learned dynamics model. Unfortunately, seminal offline MBRL methods often struggle in sparse-reward, long-horizon tasks. In this work, we introduce a novel MBRL framework, dubbed Temporal Distance-Aware Transition Augmentation (TempDATA), that generates augmented transitions in a temporally structured latent space rather than in raw state space. To model long-horizon behavior, TempDATA learns a latent abstraction that captures a temporal distance from both trajectory and transition levels of state space. Our experiments confirm that TempDATA outperforms previous offline MBRL methods and achieves matching or surpassing the performance of diffusion-based trajectory augmentation and goal-conditioned RL on the D4RL AntMaze, FrankaKitchen, CALVIN, and pixel-based FrankaKitchen.
Traditional reinforcement learning (RL)-based learning approaches for wireless networks rely on expensive trial-and-error mechanisms and real-time feedback based on extensive environment interactions, which leads to low data efficiency and short-sighted policies. These limitations become particularly problematic in complex, dynamic networks with high uncertainty and long-term planning requirements. To address these limitations, in this paper, a novel world model-based learning framework is proposed to minimize packet-completeness-aware age of information (CAoI) in a vehicular network. Particularly, a challenging representative scenario is considered pertaining to a millimeter-wave (mmWave) vehicle-to-everything (V2X) communication network, which is characterized by high mobility, frequent signal blockages, and extremely short coherence time. Then, a world model framework is proposed to jointly learn a dynamic model of the mmWave V2X environment and use it to imagine trajectories for learning how to perform link scheduling. In particular, the long-term policy is learned in differentiable imagined trajectories instead of environment interactions. Moreover, owing to its imagination abilities, the world model can jointly predict time-varying wireless data and optimize link scheduling in real-world wireless and V2X networks. Thus, during intervals without actual observations, the world model remains capable of making efficient decisions. Extensive experiments are performed on a realistic simulator based on Sionna that integrates physics-based end-to-end channel modeling, ray-tracing, and scene geometries with material properties. Simulation results show that the proposed world model achieves a significant improvement in data efficiency, and achieves 26% improvement and 16% improvement in CAoI, respectively, compared to the model-based RL (MBRL) method and the model-free RL (MFRL) method.
Reinforcement Learning (RL) has demonstrated impressive capabilities in robotic control but remains challenging due to high sample complexity, safety concerns, and the sim-to-real gap. While offline RL eliminates the need for risky real-world exploration by learning from pre-collected data, it suffers from distributional shift, limiting policy generalization. Model-Based RL (MBRL) addresses this by leveraging predictive models for synthetic rollouts, yet existing approaches often lack robust uncertainty estimation, leading to compounding errors in offline settings. We introduce Offline Robotic World Model (RWM-O), a model-based approach that explicitly estimates epistemic uncertainty to improve policy learning without reliance on a physics simulator. By integrating these uncertainty estimates into policy optimization, our approach penalizes unreliable transitions, reducing overfitting to model errors and enhancing stability. Experimental results show that RWM-O improves generalization and safety, enabling policy learning purely from real-world data and advancing scalable, data-efficient RL for robotics.
The goal of many applications in energy and transport sectors is to control turbulent flows. However, because of chaotic dynamics and high dimensionality, the control of turbulent flows is exceedingly difficult. Model-free reinforcement learning (RL) methods can discover optimal control policies by interacting with the environment, but they require full state information, which is often unavailable in experimental settings. We propose a data-assimilated model-based RL (DA-MBRL) framework for systems with partial observability and noisy measurements. Our framework employs a control-aware Echo State Network for data-driven prediction of the dynamics, and integrates data assimilation with an Ensemble Kalman Filter for real-time state estimation. An off-policy actor-critic algorithm is employed to learn optimal control strategies from state estimates. The framework is tested on the Kuramoto-Sivashinsky equation, demonstrating its effectiveness in stabilizing a spatiotemporally chaotic flow from noisy and partial measurements.
This short paper describes our proposed solution for the third edition of the "AI Olympics with RealAIGym" competition, held at ICRA 2025. We employed Monte-Carlo Probabilistic Inference for Learning Control (MC-PILCO), an MBRL algorithm recognized for its exceptional data efficiency across various low-dimensional robotic tasks, including cart-pole, ball \& plate, and Furuta pendulum systems. MC-PILCO optimizes a system dynamics model using interaction data, enabling policy refinement through simulation rather than direct system data optimization. This approach has proven highly effective in physical systems, offering greater data efficiency than Model-Free (MF) alternatives. Notably, MC-PILCO has previously won the first two editions of this competition, demonstrating its robustness in both simulated and real-world environments. Besides briefly reviewing the algorithm, we discuss the most critical aspects of the MC-PILCO implementation in the tasks at hand: learning a global policy for the pendubot and acrobot systems.
Existing visual model-based reinforcement learning (MBRL) algorithms with observation reconstruction often suffer from information conflicts, making it difficult to learn compact representations and hence result in less robust policies, especially in the presence of task-irrelevant visual distractions. In this paper, we first reveal that the information conflicts in current visual MBRL algorithms stem from visual representation learning and latent dynamics modeling with an information-theoretic perspective. Based on this finding, we present a new algorithm to resolve information conflicts for visual MBRL, named MInCo, which mitigates information conflicts by leveraging negative-free contrastive learning, aiding in learning invariant representation and robust policies despite noisy observations. To prevent the dominance of visual representation learning, we introduce time-varying reweighting to bias the learning towards dynamics modeling as training proceeds. We evaluate our method on several robotic control tasks with dynamic background distractions. Our experiments demonstrate that MInCo learns invariant representations against background noise and consistently outperforms current state-of-the-art visual MBRL methods. Code is available at https://github.com/ShiguangSun/minco.