Max Robotics — Global Service Robot Directory + US Certification

About

Matthias Humt is a researcher working at the intersection of deep learning uncertainty estimation and robotic manipulation, with a particular focus on making autonomous systems safer and more reliable. His most influential contribution is a sweeping survey on uncertainty in deep neural networks (2023), which has amassed over 1,100 citations and has become an essential reference for researchers grappling with the challenge of quantifying confidence in neural network predictions. This work systematically maps the landscape of uncertainty estimation methods across virtually every domain where neural networks are applied. Beyond his theoretical contributions, Humt applies these ideas concretely in robotics, developing systems that can grasp unknown objects with multi-fingered robotic hands under conditions of partial observability. His research on shape completion with uncertainty prediction, two-stage grasp learning architectures, and Bayesian approaches to robotic introspection collectively advance the goal of trustworthy, long-term robot autonomy. Works combining Laplace approximations, sparse Gaussian processes, and deep learning demonstrate his commitment to principled probabilistic frameworks rather than heuristic solutions. Humt's research is distinctive in bridging fundamental uncertainty theory with applied robotic challenges, making him a valuable voice for students and practitioners seeking to deploy reliable neural networks in safety-critical real-world environments.