model actuator dynamics

This work proposes a novel nonlinear dynamics modeling framework that addresses the limitations of traditional approaches—such as Euler–Lagrange equations—which are highly susceptible to measurement noise and computationally inefficient when handling nonlinear mechanical systems with external variables. By integrating a noise-robust, computationally efficient model architecture with an automated modeling pipeline, the proposed method significantly enhances inverse dynamics prediction performance. Extensive validation in representative applications, including automotive and robotic systems, demonstrates clear advantages over conventional techniques in both noise resilience and computational speed, effectively overcoming the longstanding bottleneck in modeling complex systems with strong external dependencies.

This work addresses the challenge of constructing high-fidelity actuator dynamics models from real robot data in the absence of torque sensors, current measurements, or internal motor information. The authors formulate actuator identification as a trajectory-matching problem, relying solely on joint position and velocity provided by encoders. By integrating differentiable simulation with gradient-based optimization, their unified framework accommodates a spectrum of model classes—from compact analytical formulations to neural actuator mappings—without requiring specialized test benches or torque sensing. Experimental results demonstrate that the proposed approach reduces position error from 14.20 mrad to 7.54 mrad. Furthermore, when applied to downstream motion policy training, it yields a 46% increase in robot travel distance and a 75% reduction in heading deviation.

Existing servo actuator simulations employ coarse friction models (e.g., classical Coulomb-viscous approximations), leading to poor sim-to-real transfer. This work proposes an extended friction model that systematically integrates the Stribeck effect, LuGre hysteresis, and dynamic state evolution—marking the first unified formulation enabling fully automated, trajectory-driven parameter identification from real pendulum-arm motion data. The model features a plug-and-play interface compatible with mainstream physics engines. Evaluated on four servo motors and a 2R manipulator, it reduces mean position and velocity prediction errors by 62% over baseline models and significantly improves cross-domain control policy transfer success. This work establishes a reproducible, deployable friction modeling paradigm for high-fidelity robotic dynamics simulation.

To address the sim-to-real transfer challenge in high-reduction-ratio robotic systems, this paper proposes a physics-guided gain regularization and parameter-conditioned joint training framework. Methodologically, it models PID gains as proxies for unmodeled dynamics; leverages empirically calibrated proportional gains from real-world experiments to constrain the local input–output sensitivity of an RNN-based neural controller; and enhances policy generalization via system-parameter conditioning. The approach integrates domain randomization, sensitivity analysis, and physics-informed regularization loss. Evaluated on a commercial two-wheeled balancing robot, it achieves highly consistent angular motion responses between simulation and reality—improving convergence time alignment and significantly suppressing oscillations. Crucially, the method requires neither high-fidelity dynamical modeling nor ideal (e.g., backdrivable) actuators, making it particularly suitable for low-cost, non-backdrivable hardware platforms. It quantitatively narrows the sim-to-real gap while maintaining practical deployability.

To address the susceptibility of fully actuated robot controllers to disturbances and the difficulty in simultaneously ensuring stability and high performance, this paper proposes a novel controller design framework integrating virtual mechanism modeling with passivity theory. Methodologically, the control law is physically modeled as a virtual mechanism, whose intrinsic passivity guarantees closed-loop stability; concurrently, a rigid-body dynamics ordinary differential equation model is constructed, and automatic differentiation enables end-to-end optimal tuning of controller parameters. The key contribution lies in the first deep coupling of virtual mechanism modeling with passivity analysis—overcoming the empirical parameter-tuning bottleneck inherent in conventional passive control—and achieving a unified design where stability is formally provable and performance is systematically optimizable. Simulation results demonstrate significant improvements in trajectory tracking accuracy and energy efficiency.

This study addresses the challenges of control design in complex industrial processes characterized by multivariable coupled dynamics by proposing an automated control strategy generation framework that integrates large language models (LLMs) with Bayesian optimization. The approach decomposes control design into structured code generation steps, ensuring physical consistency through execution-based validation and feedback-driven repair. It pioneers the automatic synthesis of decentralized PI controller architectures and their tuning environments directly from dynamic process models. Evaluated on a nonlinear gas preheater benchmark, the generated control schemes—subsequently refined via Bayesian optimization—achieve a 26.5% improvement in closed-loop performance and significantly enhance the transient response of pressure loops, thereby demonstrating the method’s effectiveness and novelty.

This work addresses the dynamics discrepancy between simulation and reality in parallel-legged mechanisms caused by missing actuator and link inertias due to coordinate transformations in serial tree-based simulators. To resolve this, the authors propose Simulation-side System Normalization (S3N), which preserves the serial tree topology while embedding actuator inertia and damping through coordinate transformations (S3N-Act) and further recovers link inertia via frequency response identification (S3N-Full), all without modifying the underlying simulator architecture. Experimental results demonstrate that S3N-Full reduces joint position and torque RMSE by 80.9% and 82.1%, respectively, in a 2-DOF task; decreases ground reaction force errors by over 62% during stationary pitching; and lowers the average simulation-to-reality gap in circular walking from 17.3% to 9.9%.