Almost surely safe exploration and exploitation for deep reinforcement learning with state safety estimation

K Lin, Y Li, Q Liu, D Li, X Shi, S Chen - Information Sciences, 2024 - Elsevier
This study aims to address the challenge of constrained reinforcement learning (CRL),
which seeks to maximize cumulative rewards while making agents avoid risks. Existing CRL …

Planning under uncertainty for safe robot exploration using Gaussian process prediction

A Stephens, M Budd, M Staniaszek, B Casseau… - Autonomous …, 2024 - Springer
The exploration of new environments is a crucial challenge for mobile robots. This task
becomes even more complex with the added requirement of ensuring safety. Here, safety …

Transductive active learning: Theory and applications

J Hübotter, B Sukhija, L Treven… - 38th Annual …, 2024 - research-collection.ethz.ch
We study a generalization of classical active learning to real-world settings with concrete
prediction targets where sampling is restricted to an accessible region of the domain, while …

Lipschitz safe Bayesian optimization for automotive control

J Menn, P Pelizzari, M Fleps-Dezasse… - arXiv preprint arXiv …, 2025 - arxiv.org
Controller tuning is a labor-intensive process that requires human intervention and expert
knowledge. Bayesian optimization has been applied successfully in different fields to …

Pluck and Play: Self-supervised Exploration of Chordophones for Robotic Playing

M Görner, N Hendrich, J Zhang - 2024 IEEE International …, 2024 - ieeexplore.ieee.org
Existing robotic musicians utilize detailed handcrafted instrument models to generate or
learn policies for playing because model-free or inaccurate policy rollouts might easily …

Transductive active learning with application to safe bayesian optimization

J Hübotter, B Sukhija, L Treven, Y As… - ICML 2024 Workshop …, 2024 - openreview.net
Safe Bayesian optimization (Safe BO) is the task of learning an optimal policy within an
unknown environment, while ensuring that safety constraints are not violated. We analyze …

Information-based Transductive Active Learning

J Hübotter, B Sukhija, L Treven, Y As… - arXiv preprint arXiv …, 2024 - arxiv.org
We generalize active learning to address real-world settings where sampling is restricted to
an accessible region of the domain, while prediction targets may lie outside this region. To …

Information-Theoretic Safe Bayesian Optimization

AG Bottero, CE Luis, J Vinogradska… - arXiv preprint arXiv …, 2024 - arxiv.org
We consider a sequential decision making task, where the goal is to optimize an unknown
function without evaluating parameters that violate an a~ priori unknown (safety) constraint …

Gaussian Processes in Control: Performance Guarantees through Efficient Learning

A Lederer - 2023 - mediatum.ub.tum.de
While Gaussian process (GP) models promise to enable learning control, their application
creates theoretical and practical issues. In this thesis, theoretical challenges are addressed …