Samuele Tosatto

Technische Universität Darmstadt, Universität Innsbruck

Papers

Total Citations

H-Index

About

Samuele Tosatto is a robotics and machine learning researcher whose work sits at the intersection of robot learning, probabilistic modeling, and nonparametric statistics. His research focuses on enabling robots to learn complex dynamic behaviors efficiently and reliably, with particular emphasis on inverse dynamics modeling, movement primitives, and the theoretical foundations underpinning learning algorithms. Tosatto's most influential contribution, "Learning Inverse Dynamics Models in O(n) Time with LSTM Networks" (2017, 84 citations), demonstrated that recurrent neural architectures could efficiently capture the highly nonlinear dynamics of compliant robotic systems — dynamics that elude traditional analytic models due to friction, elasticity, and sensor noise. This work has become a key reference in data-driven robot control. More recently, his research has expanded into probabilistic movement primitives, combining Bayesian aggregation with deep learning to improve generalization from limited demonstrations, a critical challenge in imitation learning. His 2020 theoretical work on Nadaraya-Watson kernel regression reflects his commitment to grounding machine learning methods in rigorous mathematics, providing finite-sample bias bounds where only asymptotic results previously existed. Tosatto's portfolio reveals a researcher bridging practical robotics challenges with principled statistical theory, making him a valuable voice for students navigating the increasingly data-driven landscape of modern robotics and reinforcement learning.