392273 Projekt: Real2Sim for Robot Manipulation (Pj) (WiSe 2026/2027)

IMPORTANT: please contact the project supervisor for participation: dmitrii.trofimov@uni-bielefeld.de

Training a robot on real hardware is slow and expensive. Training it in a
simulator is fast and cheap, but only if the simulated world resembles the real
one closely enough. Real2Sim is the attempt to build that simulated world
automatically: you photograph or film a real object, and a pipeline produces
something a physics simulator can use.

We agree on the focus in the first weeks, depending on your interests and on
what turns out to be practical. You work on a weekly basis under my supervision.

Requirements for participation, required level

Solid Python skills and the willingness to work with existing research code are
expected. An interest in 3D data, simulation, or robotics is helpful; prior
experience with any specific simulator is not required, the necessary
background is introduced in the first weeks.

Bibliography

The following surveys
give a first impression of the topic.

• E. Aljalbout et al., ''The Reality Gap in Robotics: Challenges, Solutions, and

Best Practices'', arXiv:2510.20808

• T. Ye et al., ''3D Generation for Embodied AI and Robotic Simulation: A

Survey'', arXiv:2604.26509

• A. Melnik, H. Rhodin, M. Beetz et al., survey on digital twin generation for

robotics, arXiv:2504.13159