Jan Peters

Synthetic data generation via RL not only scales but also enhances generalization in bimanual dexterous manipulation by leveraging language-conditioned task annotations.

Advanced Vision-Language-Action models can be dramatically compressed by up to 50% without losing performance, reshaping our approach to robotic manipulation.

Achieving stable five-ball juggling on robots in just two attempts reveals the critical role of directional task error in enhancing learning efficiency.

Multi-resolution tactile sensing boosts robotic manipulation success rates to 80%, far surpassing traditional vision-only approaches.

Learning dense rewards from expert demonstrations allows for over 90% success in complex manipulation tasks, outperforming traditional RL methods.

By representing deformable linear objects as a chain of relative rotations, RopeDreamer achieves state-of-the-art prediction accuracy and topological consistency in long-horizon manipulation tasks.