Dr. Manfred Eppe, Chief Engineer of the Institute for Data Science Foundations since January 2022, leaves the institute on 30 September 2026 to join the Hamburg Port Authority as AI Solution Architect.
Manfred obtained his Ph.D. in Computer Science from the University of Bremen in 2014, working on logical formalisations of symbolic reasoning, epistemic action theory and commonsense reasoning for assistive service robotics. He then worked as a postdoc at the Artificial Intelligence Research Institute (IIIA-CSIC) in Barcelona within the COINVENT project on concept representation and concept blending, and from 2015 as a postdoctoral fellow at the International Computer Science Institute (ICSI) in Berkeley on human-robot interaction and natural language understanding. From 2016 to 2021 he was a postdoctoral research associate at the Knowledge Technology group of the University of Hamburg. On 1 January 2022, he joined the Institute for Data Science Foundations as Chief Engineer.
At the institute, Manfred was responsible for the cognitive robotics lab, implemented as the freely available experimentation platform Scilab-RL (Dohmen, Röder and Eppe, arXiv:2401.14488, 2024), a software framework for reinforcement learning and cognitive modelling research on robotic agents, built on Stable Baselines 3. His research concerns computational cognitive models of representation learning and reinforcement learning for simulated and physical agents and robots, including hierarchical reinforcement learning, language grounding and contextual world models. Together with Stefan Wermter, he was principal investigator of two projects funded by the German Research Foundation (DFG): MoReSpace (DFG project 402776968, 2022–2025), within the Priority Programme SPP 2134 "The Active Self", and Learning Conversational Action Repair for Intelligent Robots (LeCAREbot, DFG grant EP 143/4-1, 2021–2024). From the summer term 2022 he taught Machine Learning I and II together with Nihat Ay. He also organised the seminar Introduction to Reinforcement Learning, co-organised the Kino-Seminar "Können Maschinen intelligent sein?" with Nihat Ay, and in the winter term 2025/2026 gave the lecture Reinforcement Learning with Frank Röder. He supervised the doctoral research of Frank Röder, who defended his thesis "Language Grounding in Deep Reinforcement Learning for Dynamic Goal-Oriented Robotics" in February 2026. On 2 April 2025, Manfred completed his habilitation at TU Hamburg. Recent work from the institute includes "Dynamics-Aligned Latent Imagination in Contextual World Models for Zero-Shot Generalization" (NeurIPS 2025 poster, presented on 4 December 2025; arXiv:2508.20294; Röder, Benad, Eppe, Banerjee) and, jointly with Nihat Ay, the preprint "Dynamics-Aligned Shared Hypernetworks for Contextual RL under Discontinuous Shifts" (arXiv:2602.06550, 2026). On a new Actuator Inversion Benchmark, the method in that preprint achieves zero-shot generalisation and outperforms domain randomisation by 58.1% on average.
At the Hamburg Port Authority, Manfred will work as AI Solution Architect on the use of artificial intelligence within the company. The Institute for Data Science Foundations thanks Manfred for almost five years of research, teaching and supervision, and wishes him every success in his new position.
Dr. Manfred Eppe, Homepage
