机器人 / 具身智能

Autonomous systems have achieved superhuman performance in isolation or simulation, yet they remain brittle in shared, dynamic real-world spaces. This failure stems from the dominant single-agent paradigm for physical applications, where other actors are ignored or treated as environmental noise, preventing effective coordination. Here we show that multi-agent reinforcement learning provides the essential safety scaffolding required for real-world interaction. Using high-speed quadrotor racing as a high-stakes testbed, we train agents to navigate complex aerodynamic interactions and strategic maneuvering with a variable number of racers. Through league-based self-play, agents evolve sophisticated anticipatory behaviors, including proactive collision avoidance, overtaking, and handling multi-agent physical interactions, including aerodynamic downwash. Our agents outperform a champion-level human pilot in multi-player races at speeds exceeding 22 m/s, while simultaneously reducing collision rates by 50 % compared to state-of-the-art single-agent baselines. Crucially, training with diverse artificial agents enables zero-shot generalization to safer human interaction. These results suggest that the path to robust robotic co-existence lies not in isolated safety constraints, but in the rigorous demands of multi-agent interaction. Multimedia materials are available at: https://rpg.ifi.uzh.ch/marl

This paper investigates a planar tracking problem between a leader and follower agent. We propose a novel feedback speed control law, paired with a constant bearing steering strategy, to maintain an abreast formation between the two agents. We prove that the proposed control yields asymptotic stability of the closed-loop system when the steering of the leader is known. For the case when the leader's steering is unavailable to the follower, we show that the system is still input-to-state stable with respect to the leader's steering viewed as an input. Furthermore, we demonstrate that if the leader's steering is periodic, the follower will asymptotically converge to a periodic orbit with the same period. We validate these results through numerical simulations and experimental implementations on mobile robots. Finally, we demonstrate the scalability of the proposed approach by extending the two-agent control law to an N-agent chain network, illustrating its implications for directional information propagation in biological and engineered flocks.

Achieving precise positioning of the mobile manipulator's base is essential for successful manipulation actions that follow. Most of the RGB-based navigation systems only guarantee coarse, meter-level accuracy, making them less suitable for the precise positioning phase of mobile manipulation. This gap prevents manipulation policies from operating within the distribution of their training demonstrations, resulting in frequent execution failures. We address this gap by introducing an object-centric imitation learning framework for last-meter navigation, enabling a quadruped mobile manipulator robot to achieve manipulation-ready positioning using only RGB observations from its onboard cameras. Our method conditions the navigation policy on three inputs: goal images, multi-view RGB observations from the onboard cameras, and a text prompt specifying the target object. A language-driven segmentation module and a spatial score-matrix decoder then supply explicit object grounding and relative pose reasoning. Using real-world data from a single object instance within a category, the system generalizes to unseen object instances across diverse environments with challenging lighting and background conditions. To comprehensively evaluate this, we introduce two metrics: an edge-alignment metric, which uses ground truth orientation, and an object-alignment metric, which evaluates how well the robot visually faces the target. Under these metrics, our policy achieves 74.58% success in edge-alignment and 89.42% success in object-alignment when positioning relative to unseen target objects. These results show that precise last-meter navigation can be achieved at a category-level without depth, LiDAR, or map priors, enabling a scalable pathway toward unified mobile manipulation. Project page: https://rpm-lab-umn.github.io/category-level-last-meter-nav/

Infants acquire language with generalization from minimal experience, whereas large language models require billions of training tokens. What underlies efficient development in humans? We investigated this problem through experiments wherein robotic agents learn to perform actions associated with imperative sentences (e.g., push red cube) via curiosity-driven self-exploration. Our approach amortizes active inference using Q-learning, enabling intrinsically motivated developmental learning. The simulations reveal key findings corresponding to observations in developmental psychology. i) Generalization improves drastically as the scale of compositional elements increases. ii) Curiosity-driven exploration enables faster learning. iii) Rote pairing of sentences and actions precedes compositional generalization. iv) Exception-handling induces U-shaped developmental performance, a pattern like representational redescription in child language learning. These results suggest that curiosity-driven active inference accounts for how intrinsically motivated sensorimotor-linguistic learning supports scalable compositional generalization and exception handling in humans and artificial agents.

Robot learning has advanced rapidly in learning control, but learning the physical body of a robot remains much more difficult because jointly searching over design and control creates a very large combinatorial problem. Here, we present a data-driven framework for generating robot hands from human demonstrations. Instead of learning a complex controller together with each candidate design, we generate robot hand designs using the same simple control policy used after fabrication: matching fingertip positions through inverse kinematics. Using more than 4 million frames of human fingertip motion from everyday manipulation, our algorithm optimizes tree-structured robot hands to reproduce desired target motions. The framework produced both a 6-degree-of-freedom (DoF) general-purpose hand and lower-DoF task-specific hands with spatial four-bar mimic joints. To accelerate the search over designs, we trained a reinforcement-learning (RL) actor to propose good hand designs and joint angles, reducing search time from hours to minutes. We fabricated the mechanisms directly as one-piece articulated structures with print-in-place joints. In real-world experiments, the 6-DoF hand achieved highly accurate teleoperated fingertip tracking better than available commercial robot hands, whereas the specialized 3-DoF hands reproduced structured human and synthetic trajectories with reduced mechanical complexity. These results showed that large-scale human motion data can be used not only to train robot controllers but also as a reference for optimizing and generating the physical embodiment of robots.

Sensor measurements are frequently corrupted by outliers and non-Gaussian noise. These imperfections in the sensor data can cause classical state estimators to generate biased and unreliable state and uncertainty estimates. Robust estimators reject or downweight outliers but do not perform measurement covariance estimation, whereas joint state and covariance estimators assume Gaussian residuals and fixed loss shape parameters. Integrating these two capabilities into a single framework is an opportunity to simultaneously estimate both state and covariance in the presence of outliers. This paper proposes a unified Block-Coordinate Descent framework that combines a norm-aware adaptive robust loss, an Iteratively Reweighted Least-Squares state update, and a Minimum Weighted Covariance Determinant covariance estimator, yielding a self-tuning joint state and covariance estimator. The framework is evaluated in a Monte-Carlo simulation and on real-world ultra-wideband localization experiments in cluttered non-line-of-sight environments. Results show that the proposed estimator consistently recovers the true inlier measurement covariance and matches or exceeds the state estimation accuracy of all baselines, without requiring any manual parameter tuning.

Actor-Critic Model Predictive Control (MPC) effectively addresses complex, non-convex control problems, but guaranteeing the closed-loop stability of sequence-based learning models within these pipelines remains challenging. This paper introduces a novel Transformer-Actor-Critic MPC architecture with formal robustness guarantees. First, we prove that Transformer networks can satisfy global incremental Input-to-State Stability ($δ$ISS). We then leverage Riemannian contraction theory to analyze the interconnected dynamics between the physical plant and the predictive neural network. Finally, we integrate these theoretical bounds as a training regularizer to yield a certifiably robust policy. The framework is validated on a nonlinear 3D drone model executing target-reaching and obstacle-avoidance maneuvers.

Dynamic environmental changes, confined workspaces, and stringent real-time constraints make pathfinding in Robotic Mobile Fulfillment Systems (RMFS) a challenging problem for conventional search- and rule-based methods, which typically suffer from high computational complexity and long decision latency. While reinforcement learning (RL) has emerged as a powerful alternative, deploying learned policies with extreme energy efficiency on resource-constrained hardware remains an open challenge. We present SDQN-RMFS, an end-to-end framework that achieves high-fidelity deployment of an RL-trained policy from a full-precision artificial neural network (ANN) through to a neuromorphic chip. By computing only when triggered by sparse events, this framework unlocks ultra-low-power RMFS pathfinding. Our full-stack pipeline operates as follows: an ANN policy is first efficiently trained via a collision-allowing strategy to densify informative trajectories, and then converted into a spiking neural network (SNN) via a hard-label knowledge distillation approach. This effectively addresses the output distribution mismatch, preserving policy capability across the ANN-to-SNN pipeline while substantially reducing inference latency. Hardware experiments demonstrate up to 11,281$\times$ energy savings and a nearly two-fold reduction in latency compared to a high-performance GPU baseline, while maintaining decision quality on par with the original trained policy. These results establish physical neuromorphic inference as a practical and energy-sustainable pathway for large-scale RMFS operations.

Humanoid robots require co-speech motions that are not only expressive and speech-aligned, but also physically executable under embodiment constraints. Existing co-speech generation pipelines are predominantly human-centric: motions are first generated in human-body representations such as SMPL-X and subsequently retargeted to humanoid robots. In this work, we identify a fundamental embodiment gap in this paradigm, where the mismatch between human motion manifolds and humanoid embodiment constraints disrupts embodiment consistency during motion transfer and physical execution. Through extensive analysis, we show that although retargeting can preserve coarse motion semantics, it significantly compresses motion diversity and weakens prosody-motion synchronization, limiting expressive humanoid behaviors. To address this problem, we first propose IK-EER, a prosody-preserving humanoid motion curation framework that jointly optimizes kinematic feasibility and speech-motion temporal alignment during retargeting. Building upon the curated robot-native motion dataset, we further introduce PhysDrift, an embodiment-aware co-speech motion generation framework that directly predicts executable humanoid joint trajectories from speech without relying on intermediate human-body representations. Unlike conventional human-centric pipelines, PhysDrift maintains embodiment consistency throughout both training and inference while incorporating physical regularization to stabilize robot motion dynamics. Extensive experiments and real-world humanoid deployment demonstrate that embodiment-aware robot-native generation substantially improves speech-motion alignment, physical plausibility, motion smoothness, inference efficiency, and real-time interaction capability.

Art has long stood as a pivotal expression of human creativity. Embodied artificial intelligence offers a route for generative models to participate in that creativity through physical action rather than disembodied digital content. In robotic music co-creation, it is challenging to connect semantic musical understanding with real-time and physically executable performance. We present Co-policy, a framework for human-robot musical co-creation that separates semantic intent grounding, constrained musical variation, and visuomotor execution. To ground musical semantics, Co-policy uses pre-inference semantic anchors and a fine-tuned Qwen-vl planner (F-Qwen) to transform speech, live musical seeds, and visual observations into structured co-creation plans. To support low-latency execution, Co-policy introduces a Gaussian-Mixture Visuomotor Policy (GMP), implemented as a conditional mixture-density policy that maps target notes and visual context to multimodal robot actions in a single forward pass. Unlike robotic playback systems that merely reproduce user-specified notes, Co-policy generates complementary musical responses under both musical and physical constraints. Real-robot chime experiments, ablations, and expert evaluation show improved intent alignment, execution accuracy, and response frequency over diffusion-policy and ablated baselines, supporting physically grounded action generation as a key requirement for embodied human-AI co-creation.

Field-based modeling from onboard measurements can produce autonomous underwater vehicle (AUV) maneuvering models that reflect real operating characteristics. From an approximation perspective, conventional maneuvering models use predefined constraint polynomial bases, whereas data-driven models use data-adaptive bases. Motivated by this basis-function view, this paper presents a differentiable composite-approximation formulation, in which the polynomial-basis component and the data-adaptive basis component are treated as differentiable parts of a single predictor and calibrated jointly. A gradient-based co-calibration method is developed for full-scale AUV maneuvering prediction, where a sensitivity-aware mechanism regulates bounded polynomial updates while the neural residual captures remaining nonlinear discrepancies under a shared prediction objective. To account for ocean-current effects in field data, a turning-motion-based current estimation and compensation procedure is incorporated to construct current-compensated learning targets for training and rollout. The framework is evaluated using sea-trial data collected from a 7-meter AUV under multiple maneuvering conditions. Results show that the proposed method improves recursive trajectory and velocity prediction compared with polynomial-only, neural-only, and frozen-prior hybrid baselines, demonstrating its applicability to field-data-based AUV maneuvering modeling.

Legged robots are increasingly deployed in forests for ecological surveying and monitoring, yet their autonomy is often interrupted consequent to the challenges posed in traversing forest environments. Forest entrapments, for example, when a robot's legs are ensnared in vines or other vegetation, result in loss of stability and toppling. Such events not only disrupt the mission and require manual intervention, but also risk damage to the robot hardware. To address the absence of a dedicated dataset to investigate these failure modes in forest environments, we present ForEnt, a multi-modal dataset collected with the low-cost Unitree Go2 quadruped across eight forest sites in the Southampton Common Woodlands, UK. For our dataset, over approximately 1.7 km of traversals in 11 sequences were conducted, yielding 69 recorded entrapment events. ForEnt includes time-synchronized RGB-D images, LiDAR scans, proprioceptive data, and third-person video, enabling analysis of terrain factors contributing to entrapment and providing labeled sensor streams for reproducible benchmarking. By supporting the evaluation of entrapment detection strategies, ForEnt lowers the barrier to developing robust quadruped robot deployments in challenging forest environments.

A natural recipe for intelligent robotic decision-making is initializing from pretrained generative control policies, which have summarized offline experience, and adapting them to self-collected online experience. We present DF-ExpEnse, an exploration technique that improves the quality of online experience collection, thus increasing finetuning sample-efficiency. DF-ExpEnse leverages the multimodal modeling capabilities of the generative control policy to create an expressive and tractably evaluatable candidate set. It then utilizes an ensemble of critics to identify the action that best balances quality with high exploration interest. In fleet settings, DF-ExpEnse further enables cross-agent communication to facilitate collaborative exploration as a group. DF-ExpEnse can be seamlessly integrated with existing strategies that finetune pretrained generative control policies via reinforcement learning. We experimentally validate consistent sample-efficiency benefits through DF-ExpEnse across a variety of manipulation and locomotion tasks, compared to default finetuning and alternative action selection schemes. Project can be found at https://df-expense.github.io.

Multi-agent reinforcement learning (MARL) enables agents to develop coordination strategies through emergent communication, but neural policies lack the formal safety guarantees required for safety-critical robotic deployment in drone swarms and autonomous vehicle fleets. We present the first end-to-end framework for safety verification of learned multi-agent communication policies through policy abstraction: neural policies are distilled into interpretable decision trees, then formally verified, with empirical validation confirming that verified safety properties transfer to original networks. Our four-stage pipeline consists of domain-specific feature extraction from agent observations, decision tree distillation achieving 97.9% +/- 1.2% fidelity to neural policies, automated translation to PRISM probabilistic model checker specifications with complete feature-to-state-variable correspondence, and compositional verification of Probabilistic Computation Tree Logic (PCTL) properties via pairwise decomposition with union-bound aggregation and empirical neighbor modeling. Evaluating Vector-Quantized Variational Information Bottleneck (VQ-VIB) policies for multi-drone coordination with 5-7 agents, we verify 18 temporal logic properties across safety, liveness, and cooperation, achieving 88.9% property satisfaction with all five safety thresholds satisfied (0.3% collision probability vs. 1% threshold). Monte Carlo validation of original neural policies confirms that verified safety properties transfer with <=0.6 percentage-point deviation (95% CI). Discrete VQ-VIB messages provide +11.6 to +13.6 percentage-point fidelity advantages over continuous methods, enabling 3-4x faster verification. Our framework provides empirically validated safety verification for distilled policy abstractions, serving as a practical bridge between deep MARL and formal safety workflows for multi-robot deployment.

This paper presents an invariant extended Kalman filtering (InEKF) approach for real-time state estimation of humanoid robots operating on non-inertial ground using only onboard proprioceptive sensing. The proposed approach estimates the robot's base position and velocity relative to the moving ground frame without requiring direct measurements of ground motion or externally mounted sensors. By exploiting kinematic constraints at the stance foot through foot-mounted IMUs, the filter accounts for ground-induced nonlinearities in the process and measurement models while remaining fully proprioceptive. The estimator is formulated to admit a right-invariant measurement model, enabling favorable error dynamics under large initial uncertainties. Observability analysis establishes conditions under which the robot's relative base position and velocity are observable with respect to the non-inertial ground frame. Experiments with the Digit humanoid robot standing and squatting atop a swaying and pitching ground showcase a 96% speedup in convergence rate and an 80% reduction in position estimate errors over existing InEKFs. Walking experiments on a uni-axially rotating ground achieve an average estimation error of less than 9 cm for an initial error of up to 1 m.

Autonomous mobile service robots are often required to complete tours that require navigating through a set of locations in an environment. Example domains include guiding people through a shopping mall, delivering packages in a fulfilment centre, or giving guided tours in a museum. However, in crowded environments, the presence of people may negatively impact robot performance. For example, humans will activate robot collision avoidance manoeuvres that slow the robot down. Crowds move stochastically and vary throughout the day. In this paper we present a probabilistic tour planner for crowded environments which explicitly reasons over human congestion. We learn circular linear flow field (CLiFF) maps which predict human trajectories given an initial observation. We then use these predictions to build and solve a Markov decision process online which efficiently routes the robot through the environment. Our approach is scalable enough to re-plan as new people are observed. We evaluate our approach on a real-world crowd dataset in a shopping mall.

Recent advancements in Resistive Force Theory (RFT) enable approximation of ground reaction forces for locomotion in sand without the computational expense of modeling interactions with individual grains. However, these tools have been absent in 3D physics engines commonly used for robot simulation. We explore if resistive force approximations are sufficient, when integrated with standard dynamics calculations, to provide a stable substrate for a freely walking robot. To determine this, we implement 3D Granular Resistive Force Theory (3D RFT) in a physics simulation engine, MuJoCo. We verify simulations in multiple scenarios to demonstrate that key trends due to end effector shape, speed, and loading are preserved. Our implementation predicts walking distance and foot sinkage of a 12-Degree of Freedom hexapod robot within 20\% of experiments in sand. While RFT has inherent approximations, the open source tool described here has potential to help develop new and improved robot designs to traverse granular media substrates.

Latent actions provide a compact interface between action-free video and downstream decision-making, yet existing Latent Action Models (LAMs) force every transition through a fixed-capacity bottleneck. We identify a bottleneck trade-off: overly tight codes can discard transition cues needed for action alignment, while overly loose codes preserve additional transition variation that must be resolved when alignment labels are scarce or narrowly distributed. FlexLAM replaces this fixed capacity with variable-length latent actions trained by nested dropout, yielding prefix-valid codes that capture compact transition structure first and add detail only when needed, without new architectures or losses. A single FlexLAM matches or surpasses separately trained fixed-capacity LAMs at every evaluated token budget under standard scarce-label supervision and under a low-return single-task alignment stress test, indicating that FlexLAM is not merely adjustable at inference time but learns a better latent-action interface at the same token budgets. The same model supports inference-time token-budget adjustment without retraining, and FlexLAM improves Ego4D transition reconstruction. These results suggest that variable-length latent actions are an architecture-free, drop-in upgrade to the fixed-capacity bottleneck in latent action models, latent-action world models, and video-pretrained action interfaces.

Karst aquifers provide critical freshwater resources but pose significant hazards due to their complex and poorly understood subsurface geometry. Mapping these environments is challenging because sonar data from underwater exploration is sparse and noisy, while navigation estimates suffer from drift limiting standard 3D reconstruction methods. We present a pipeline for reconstructing underwater karst conduits from a sonar profiler. We combine a continuous-time SLAM approach to correct trajectory drift with a novel two-stage deep learning method for surface reconstruction, producing an immersive and navigable 3D mesh for hydrogeological analysis.

AI vision models are a driving factor for the potential use case scenarios of cognitive robotics within in the industry and household applications. A large array of methods from semantic environment analysis towards 6D and grasping pose estimation have been proposed based on the latest AI achievements. However, such advancements require further strong and efficient methods w.r.t. training data and AI-architectures, which are capable in synergy to tackle current challenges, precision limits, and scalability beyond domain gaps. In this paper, we discuss these current limits and trends in the related state-of-the-art which are challenging those. Further we discuss our current work in progress on bridging the domain gap between simulations and real world applications by linking those in the training data generation.

Vision-Language-Action (VLA) models pre-trained on massive video-robot datasets have revolutionized robotic manipulation, yet their multi-billion parameter architectures impose prohibitive computational burdens during downstream fine-tuning and real-time inference. In this work, we reveal a highly non-trivial architectural characteristic of these continuous control foundation policies (e.g., pi_0, GR00T-N1.5): despite being trained on diverse physical trajectories, they exhibit severe layer-wise representational redundancy. To exploit this, we introduce a structural compression pipeline that is entirely training-free, bypassing the need of existing methods to load full-scale models to learn optimized token reductions or dynamic layer selectors. Instead, using only a single forward pass via Centered Kernel Alignment to identify redundant layer features, we remove twin layers to permanently compress the model depth by up to 50% across both the VLM backbone and the continuous control policy head. Downstream fine-tuning of this streamlined architecture yields a dual acceleration benefit: a 40-50% reduction in training time and up to 30% faster real-time inference, while matching or exceeding full-scale base model performance. We comprehensively validate our method across three simulation benchmarks (LIBERO, RoboCasa, SimplerEnv) and 10 diverse real-world manipulation tasks across 4 unique robotic embodiments. These results prove that advanced VLAs require significantly fewer layers than previously assumed, offering a highly compute-efficient paradigm for scalable robot learning.

Scaling robot policy learning for autonomous surgery is challenging, as expert demonstrations are expensive and in vivo exploration poses substantial safety risks. Surgical world models address this by generating realistic, action-conditioned future frames from an initial observation, but existing methods exhibit two persistent failure modes: spatial interaction incoherence, where visible instrument contact fails to induce spatially consistent tissue deformation, and temporal fidelity collapse, where prediction errors compound across autoregressive rollouts and progressively corrupt visual quality. We present SurgVista, a surgical world model that mitigates both failures through two training recipes. Deformation Consistency Regularization extracts scene-point trajectories from training videos and enforces cross-frame coherence through latent contrastive learning, strengthening physically consistent instrument-tissue dynamics. Drift Adaptation Training mitigates long-horizon drift by perturbing conditioning frames with online prediction residuals and photometric augmentations calibrated to long-horizon drift statistics, sustaining visual fidelity over extended rollouts. To enable rigorous evaluation, we further introduce SurgWorld-Bench, featuring diverse procedure types, long-range rollouts, and decoupled metrics for instrument-motion accuracy and tissue-response fidelity. Extensive experiments show that SurgVista consistently outperforms state-of-the-art methods across visual quality, temporal consistency, and interaction fidelity, with gains widening as the prediction horizon grows.

In safety-critical scenarios, the protection level of the autonomous navigation system is crucial for enabling mobile robots to perform safe tasks. However, existing studies on probabilistic navigation systems for robots usually perform offline accuracy evaluations using limited datasets and assume that the results can be applied to unknown real-world environments. As a result, current autonomous mobile robots often lack protection levels for online safety assessment. To fill this gap, we propose a safety-critical LiDAR-inertial odometry (LIO) that provides deterministic protection levels based on on-manifold deterministic state estimation. By adopting the unknown but bounded assumption, we derive a neat closed-form relationship between point cloud noise and the uncertainty of the estimation from the iterated closest point algorithm. Using this relationship, we design an on-manifold ellipsoidal set-membership filter and implement it within the LIO system. Leveraging the properties of the set-membership filter, our system offers the feasible sets of the estimated locations as the deterministic protection levels, serving as safety references for the robots' downstream autonomous operations. The experimental results show that our system can provide effective deterministic online safety references for diverse robots in various environments.

Controlling flapping-wing drones requires controllers that handle time-varying, nonlinear, underactuated dynamics from incomplete, noisy sensor data. Recent advances in artificial intelligence (AI), particularly reinforcement learning (RL), have opened new perspectives for addressing such complex control problems through data-driven policy optimization from interaction with the environment. Yet purely data-driven methods are sample-inefficient, demanding extensive, sometimes unsafe exploration, especially without guiding physical models. This motivates hybrid AI-physics frameworks. This article proposes a hybrid model-free/model-based flight-control approach using the reinforcement twinning algorithm. The model-based (MB) component uses an adjoint formulation and an adaptive digital twin continuously identified from live trajectories; the model-free (MF) component uses RL. The two agents share knowledge via transfer learning, imitation learning, and shared experience between the real environment and the digital twin, coordinated by a policy referee that selects which agent acts in reality based on digital-twin performance and a real-to-virtual consistency ratio. The framework is evaluated for the longitudinal control of a flapping-wing drone, modelled as a nonlinear time-varying system driven by quasi-steady aerodynamic forces. The hybrid strategy is tested under three adaptive-model initializations: (1) offline identification from existing data, (2) random initialization with fully online identification, and (3) offline pre-training with biased parameters followed by online adaptation. In all cases, the hybrid framework improves performance, robustness, and sample efficiency over purely model-free and purely model-based approaches.

In challenging real-world reinforcement learning applications, communication delays or sensor failures often cause frame dropping, in which the agent cannot receive the dropped states and associated rewards. To address the performance degradation caused by frame dropping, the Decision Transformer under Random Frame Dropping (DeFog) was developed by incorporating additional mechanisms into the decision transformer to tackle frame dropping. Although DeFog can mitigate performance degradation in frame-dropping environments, since DeFog is an offline learning method, it struggles to effectively generalize to novel states not adequately represented in the training dataset. In this study, we propose OnDeFog, which integrates the mechanisms in DeFog with the online decision transformer (ODT), an online reinforcement learning method that learns policies through direct environmental interaction. Comprehensive experimental evaluation demonstrates that our proposed OnDeFog achieves superior performance compared to ODT in environments characterized by high dropping frame rate and outperforms DeFog on datasets containing a large amount of low-reward data.