Tag
一篇探讨深度强化学习在高超声速/高速流动控制中应用前景与挑战的综述性论文，讨论了可解释性、仿真保真度、数据效率及工程落地路径等关键约束。
This paper systematically studies the sample-efficiency of Kolmogorov-Arnold Networks (KANs) versus MLPs on the Feynman dataset and Gymnasium RL benchmarks, showing 40% fewer samples needed and up to 50% relative performance gains during training, robust to reward noise.
The article explores the challenges of controlling AI coding agents after code changes, proposing an engineering control plane for permissions, verification, and human approval.
The article discusses the growing disparity between AI agent capabilities and the necessary control mechanisms for production use, highlighting challenges in permissions, escalation, and accountability.
The paper proposes a data-driven framework for feedback linearization control using neural Lie derivatives and augmented Lagrangian, with theoretical stability guarantees validated on a DC motor system.
This paper proposes a two-step strategy to correct learning-based perception errors in autonomous systems, using offline uncertainty characterization and a runtime risk heuristic to improve safety with minimal impact on performance.
D-JEPA is a decision-aligned latent world model that learns decision-relevant relations among candidate futures to improve action selection in tasks like manipulation and autonomous driving, demonstrating significant performance gains.
This paper proposes a composite-gradient learning method that integrates deep reinforcement learning and model predictive control for shared control authority in autonomous systems, with evaluations on traffic networks showing modest benefits under strong interaction.
This paper establishes generator-independent runtime assurance theorems, showing necessary and sufficient conditions for safety guarantees in systems using black-box generators under partial observation, with applications to learning-enabled control.
The paper proposes NOMAD-RL, a reinforcement learning controller for HVAC systems that uses adaptive domain randomization to generalize across heterogeneous thermal zones, approaching the performance of model predictive control.
This paper studies Action Diffusion, a conditional diffusion model for generating action sequences in open-loop control systems with dry friction and stiction, showing it reduces terminal error and stuck steps compared to other methods.
This paper explores the irony that automation in systems can lead to deskilling of human operators and increased workload, highlighting the need for human-centered design in automated environments.
This paper argues that the machine intelligence architectures proposed by Edmund C. Berkeley and David L. Heiserman remain relevant for embodied robotic cognition, emphasizing logical states, hardware-near control, and modular sensorimotor structure.
A novel generalist controller using attention mechanisms and mixture-of-experts is proposed, enabling a single neural network to control diverse dynamical systems without system-specific tuning. It achieves comparable performance to traditional controllers across 25 different systems.
This work-in-progress paper proposes embedding a differentiable physics model into the PPO actor loss function to penalize anticipated safety violations in reinforcement learning, evaluated on a simulated 1-DoF helicopter system. The physics-informed soft regularizations reduce constraint violations while maintaining reliable target tracking.
University of Michigan Robotics shares free open-source course materials including lectures, textbooks, and projects from their top robotics program, covering topics from computational linear algebra to autonomous systems.
This paper investigates memory-efficient meta-reinforcement learning architectures for adaptive safety-critical control in adversarial spacecraft proximity operations, finding that state space models like Mamba with PPO achieve superior task completion, safety, and fuel savings compared to LSTM and GRU.
Autonomous coding is evolving from better prompting to better control systems, with engineers wrapping agents in goals, evaluators, and loops.
This paper proposes PE-MHL, a Physics-Encoded Modular Hybrid Layer framework that incrementally refines a physics-based model with data-driven sub-models, providing theoretical convergence guarantees and outperforming monolithic networks on control benchmarks.
A unified Python framework using PPO-based deep reinforcement learning for optimizing HVAC control with economizer logic and CO2-constrained ventilation is presented, showing improved energy efficiency and temperature stability over traditional PID controllers.