sim-to-real transfer

Model mismatch in sim-to-real transfer severely hinders deployment of robotic policies trained in simulation. Method: This paper proposes the Real-Sim-Real (RSR) closed-loop framework, which jointly optimizes policy and simulator parameters via gradient-based online system identification using differentiable simulation (MuJoCo MJX), dynamically aligning simulated dynamics with real-world behavior. An information-theoretic cost function guides active, diverse real-world data collection to maximize parameter identifiability. Unlike conventional unidirectional sim-to-real pipelines, RSR establishes an iterative “real → sim → real” optimization loop, tightly coupling reinforcement learning (PPO/SAC) with online system identification. Contribution/Results: RSR significantly reduces the sim-to-real performance gap across diverse manipulation tasks. It demonstrates strong robustness to both explicit (e.g., mass, friction) and implicit (e.g., unmodeled contact dynamics) environmental uncertainties, and exhibits superior cross-scenario generalization without task-specific retraining.

Bridging the domain gap between real-world RGB-D images and robot simulation environments remains challenging for digital twin task generation. Method: This paper proposes a simulation-task alignment framework leveraging vision-language models (VLMs) and an iterative routing mechanism to generate executable simulation tasks end-to-end from single-frame RGB-D input. The method integrates SAM2 for precise object segmentation, VLM-driven semantic understanding, dynamic matching against a simulation asset library, and automated generation of self-validating test suites—forming a closed-loop “perceive–match–generate–verify” optimization pipeline. Contribution/Results: It achieves the first high-fidelity geometric-semantic alignment between real-scene objects and simulation assets while ensuring physical feasibility and executability within physics engines. Evaluated on multiple real-world benchmarks, the approach significantly improves object correspondence accuracy (+23.6%), task success rate (+31.4%), and cross-scene generalization.

High-speed极限 maneuvering of wheeled robots suffers from time-varying tire friction parameters (e.g., due to tire degradation), model mismatch, and slow online adaptation. Method: This paper proposes an adaptive control framework integrating meta-learning pretraining with uncertainty-aware Model Predictive Path Integral (MPPI) control. Leveraging Bayesian neural networks for dynamics modeling and probabilistic uncertainty quantification, the method achieves rapid cross-platform and cross-parameter adaptation using only minimal online data—without task- or robot-specific re-tuning. Results: Evaluated in numerical simulation, large-scale Unity simulation, and on a mid-sized physical robot platform, the approach matches the performance of hand-tuned controllers while significantly enhancing robustness and generalization against unknown dynamics and time-varying friction characteristics.

A significant gap exists between academic research and industrial practice in robot runtime software reconfiguration: while the literature predominantly focuses on structural reconfiguration (e.g., component loading/unloading), industry widely adopts lightweight, low-intrusion parameter-based reconfiguration. Method: We conduct a mixed-methods study comprising a systematic literature review of 78 papers, source-code and documentation analysis of four mainstream robotic frameworks, and empirical evaluation across 48 subsystems plus real-world case studies. Contribution/Results: This work is the first to systematically expose this practice-research divide and establishes the first comprehensive design space for robot reconfiguration. It confirms parameter-level reconfiguration as the sole widely adopted paradigm in practice; identifies critical research gaps and emerging technical trends; and proposes actionable, engineering-oriented improvement pathways. Our findings provide empirically grounded guidance for practitioners in tool selection, framework design, and industry–academia collaboration.

To address degraded control accuracy in sim-to-real transfer caused by dynamical mismatches, this paper proposes a gradient-free in-context learning method: leveraging the robot’s historical interaction sequence as dynamic prompts to adaptively tune simulation parameters online for rapid and precise system identification. This work is the first to introduce in-context learning into sim-to-real system identification, integrating a Transformer architecture, a gradient-free online adaptation mechanism, and dynamics-deviation-aware prompt encoding—thereby eliminating reliance on domain randomization and differentiable optimization. Evaluated on object scooping and tabletop pneumatic air-hockey tasks, the method reduces simulation parameter estimation error by 80% and 42%, respectively, and achieves ≥70% scooping success across three distinct object categories.

This work addresses the performance degradation of locomotion policies in quadrupedal robots during sim-to-real transfer caused by dynamics mismatches. To bridge this gap, the authors propose a simulator adaptation method based on proprioceptive distribution matching, which aligns the distributions of observations and actions between simulation and real hardware without requiring temporal alignment, external sensing, motion capture, or precise initial conditions. The approach jointly optimizes simulator dynamics parameters through parameter identification, an action delta model, and a residual actuator model to achieve efficient adaptation. Experiments on the Go2 robot demonstrate that with less than five minutes of real-world data, the method substantially reduces trajectory drift and significantly enhances policy performance—even enabling challenging bipedal walking tasks.

This work proposes a novel “betting”-based methodology for efficiently and accurately evaluating robotic performance in real-world settings where physical experimentation is constrained. By introducing betting theory into sim-to-real performance assessment—a first in the field—the approach constructs an estimator theoretically superior to Monte Carlo estimation. The method integrates control variate approximation, cross-fidelity simulation, and statistical decision rules to enable practical deployment. Experimental results demonstrate its efficacy on synthetic data and simulated environments, and it successfully infers real-world robotic grasping accuracy with significantly fewer physical trials while improving evaluation precision.

Simulation-to-real transfer remains a central challenge in robotics, as mismatches between simulated and real-world dynamics often lead to failures. While reinforcement learning offers a principled mechanism for adaptation, existing sim-to-real finetuning methods struggle with exploration and long-horizon credit assignment in the low-data regimes typical of real-world robotics. We introduce Simulation Distillation (SimDist), a sim-to-real framework that distills structural priors from a simulator into a latent world model and enables rapid real-world adaptation via online planning and supervised dynamics finetuning. By transferring reward and value models directly from simulation, SimDist provides dense planning signals from raw perception without requiring value learning during deployment. As a result, real-world adaptation reduces to short-horizon system identification, avoiding long-horizon credit assignment and enabling fast, stable improvement. Across precise manipulation and quadruped locomotion tasks, SimDist substantially outperforms prior methods in data efficiency, stability, and final performance. Project website and code: https://sim-dist.github.io/

This work addresses the performance degradation commonly observed in sim-to-real transfer of reinforcement learning for robotic navigation, which stems from domain discrepancies and a lack of systematic analysis linking training strategies to deployment outcomes. The authors propose an end-to-end training and deployment pipeline that decouples key influencing factors, introducing perturbation-aware fine-tuning and a Transformer-based temporal reasoning policy to significantly enhance zero-shot transfer robustness and control smoothness. By integrating perturbation modeling, post-training fine-tuning, and system-level domain gap analysis, the method outperforms existing learning-based baselines in both static and dynamic environments, matching the performance of optimization-based planners in static scenes and achieving successful zero-shot deployment across multiple real-world robotic platforms.

This work addresses the sim-to-real transfer failure commonly encountered in reinforcement learning due to mismatches between idealized actuator models used in simulation and the nonlinear, hardware-dependent motor dynamics of real robots. To bridge this gap, the authors propose “actuator reality shaping,” a method that deploys a two-degree-of-freedom feedforward–feedback controller on physical hardware to shape the closed-loop actuator response to closely match an ideal second-order reference model assumed in simulation. Notably, this approach requires no system identification or learned actuator models and enables zero-shot policy deployment through a standardized actuator interface. Experiments across diverse platforms—including single-joint servos, a 7-DoF manipulator, wheeled-legged robots, and humanoids—demonstrate substantial reductions in tracking error and successful zero-shot transfer across multiple tasks and systems, thereby shifting the paradigm from increasing simulation fidelity to unifying real-world actuator behavior to conform to simulation assumptions.