Institution profile

German Research Center for Artificial Intelligence

Academic institutioneurope · de
Official website
Research library699linked papers
Opportunities0open roles
Selected work

Representative Papers

A Survey of Reinforcement Learning from Human Feedback

Dec 22, 2023arXiv.org

Reinforcement learning (RL) often relies on hand-crafted reward functions that struggle to align with complex, nuanced human values. Method: This work systematically reviews RL from Human Feedback (RLHF), integrating reinforcement learning, Bayesian inference, preference modeling, reward modeling, and human-in-the-loop evaluation to support heterogeneous, multi-source feedback. It introduces the first unified, cross-task and cross-modal analytical framework for RLHF—extending beyond traditional preference-based RL (PbRL) limitations. Contribution/Results: The framework establishes a rigorous theoretical foundation and practical roadmap for human-AI value alignment. It clarifies the technical evolution, identifies core challenges (e.g., feedback sparsity, bias propagation, scalability), and proposes a standardized taxonomy for RLHF research. This serves as a comprehensive guide for algorithm design, ethical assessment, and real-world deployment—enabling principled, scalable, and value-aligned AI systems.

263 citations11 influentialRead paper

DMH-HARQ: Reliable and Open Latency-Constrained Wireless Transport Network

Dec 07, 2022

Achieving ultra-reliable, low-latency end-to-end (E2E) communication in 6G multi-hop wireless networks under finite blocklength (FBL) constraints remains challenging. Method: This paper proposes a Dynamic Multi-hop Hybrid ARQ (DMH-HARQ) mechanism. It introduces the first FBL-based analytical model and joint optimization of E2E reliability for multi-hop decode-and-forward (DF) relaying. A resource allocation algorithm based on integer dynamic programming is designed, integrating network virtualization and the O-RAN open architecture to enable dynamic scheduling. Contribution/Results: DMH-HARQ overcomes the performance limitations of conventional static HARQ and listen-before-talk cooperative ARQ schemes. Under strict latency constraints, it significantly improves E2E reliability—especially in long-distance, multi-hop scenarios—without requiring additional delay adaptation. Moreover, it natively supports open RAN architectures, ensuring seamless integration with emerging 6G network designs.

2 citationsRead paper

LIEREx: Language-Image Embeddings for Robotic Exploration

Jan 30, 2026KI - Künstliche Intelligenz

This work addresses the limitation of traditional semantic mapping, which relies on predefined object categories and struggles to handle unknown objects, thereby hindering goal-directed exploration in partially unknown environments. To overcome this, the authors propose an open-vocabulary semantic mapping approach that integrates vision-language foundation models—such as CLIP—with 3D semantic scene graphs. This method introduces open-vocabulary semantic embeddings into 3D scene graph construction for the first time, effectively bypassing the constraints of fixed taxonomies. By enabling natural language–guided exploration strategies, the framework facilitates robust recognition and semantic reasoning about out-of-distribution target objects, significantly enhancing the robot’s semantic understanding and task generalization capabilities in dynamic and unfamiliar settings.

1 citationsRead paper

ExPrIS: Knowledge-Level Expectations as Priors for Object Interpretation from Sensor Data

Jan 21, 2026KI - Künstliche Intelligenz

This work proposes an expectation-guided dynamic semantic scene understanding framework to address the lack of semantic consistency in existing purely data-driven robotic object recognition methods, which struggle to incorporate environmental priors. By constructing a 3D semantic scene graph that integrates contextual priors with external knowledge bases such as ConceptNet, the framework introduces knowledge-level expectations as a prior for interpreting sensor data. Object reasoning is performed within a heterogeneous graph neural network, leveraging these knowledge-driven expectations to guide perception. The approach significantly enhances both semantic consistency and temporal continuity of object recognition in dynamic environments, enabling robots to interpret scenes more coherently over time.

1 citationsRead paper

Enhancing Robustness of Asynchronous EEG-Based Movement Prediction using Classifier Ensembles

Jan 07, 2026arXiv.org

This study addresses the challenge of high false-positive rates and noise susceptibility in online asynchronous EEG-based motor intention detection, which undermines the reliability of rehabilitation robot triggering. For the first time, it systematically validates the efficacy of classifier ensembles in asynchronous EEG decoding by integrating SVM, MLP, and EEGNet into a unified ensemble model, complemented by a sliding-window post-processing mechanism. In both offline and pseudo-online evaluations, the proposed approach significantly outperforms the best individual classifier: it effectively suppresses early false alarms and achieves superior robustness and accuracy under reasonable window configurations, with particularly pronounced gains in simulated online scenarios.

1 citationsRead paper
Recent publications

Latest Papers