Institution profile

University of Klagenfurt

Academic institutioneurope · at
Official website
Research library117linked papers
Opportunities0open roles
Selected work

Representative Papers

Autonomous Control of Redundant Hydraulic Manipulator Using Reinforcement Learning with Action Feedback

Oct 23, 2022IEEE/RJS International Conference on Intelligent RObots and Systems

Hydraulic-driven redundant manipulators face challenges in autonomous control due to complex system modeling and strong reliance on precise dynamic parameters. Method: This paper proposes an end-to-end data-driven approach requiring only minimal simulation priors and teleoperated demonstration data. It employs Actuator Networks to model nonlinear hydraulic dynamics and integrates forward-kinematics–guided supervision into a modified DDPG framework—enhanced with Ornstein–Uhlenbeck noise for exploration—to directly output joint-level commands for 3D end-effector pose tracking. Contribution/Results: We introduce, for the first time, kinematic feedback within the RL action-selection mechanism, eliminating the need for system identification, inverse-dynamics modeling, or post-deployment fine-tuning. Evaluated on a scaled 3R1P hydraulic logging crane, the policy trained purely in simulation transfers zero-shot to hardware, achieving high-precision 3D position tracking. This significantly advances the feasibility and robustness of data-driven control for strongly nonlinear hydraulic systems.

5 citationsRead paper

Large Deviations of Gaussian Neural Networks with ReLU activation

May 27, 2024arXiv.org

This work investigates the large-deviation behavior of deep neural networks with Gaussian i.i.d. weights under linearly growing (unbounded) activation functions—specifically ReLU. Addressing the limitation of existing large-deviation theory, which applies only to bounded continuous activations, we establish the first rigorous large-deviation principle for the ReLU case. Our method integrates tools from random matrix theory, Gaussian process analysis, and power series expansions to derive a concise, closed-form rate function. Crucially, we obtain an explicit power series representation of this rate function tailored to ReLU. The theoretical results align closely with empirical observations in modern deep learning architectures. This framework provides a novel analytical tool for quantitatively characterizing neural network generalization, informing principled weight initialization schemes, and elucidating training dynamics—thereby advancing the theoretical foundations of deep learning.

1 citations1 influentialRead paper
Recent publications

Latest Papers