Iman Nematollahi
Papers
7
Total Citations
33
H-Index
3
About
Iman Nematollahi is a robotics researcher whose work spans robot learning, skill acquisition, and video prediction, with a particular focus on enabling autonomous agents to adapt and generalize across complex, real-world tasks. His research sits at the intersection of reinforcement learning, imitation learning, and world models, addressing one of robotics' most persistent challenges: how robots can learn robust, transferable skills without requiring prohibitive amounts of training data. Among his most notable contributions is the development of Soft Actor-Critic Gaussian Mixture Models for robot skill adaptation and generalization, work that has attracted 14 and 2 citations respectively, demonstrating growing community interest. His T3VIP framework advanced 3D video prediction by explicitly modeling spatial world dynamics, while his more recent LUMOS framework introduces language-conditioned imitation learning that leverages world models for zero-shot transfer to physical robots. His Bayesian optimization approach to manipulation learning further reflects a consistent commitment to sample efficiency—a critical bottleneck in practical robotics deployment. Nematollahi's earlier work on unsupervised structured dynamics models and non-geometric feature augmentation in learning from demonstration shows a research trajectory that has steadily grown in ambition and sophistication, making him an emerging voice in robot learning research.
Research Focus
Key Achievements
Top Papers
- 1Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models14 citations · 2022
- 2T3VIP: Transformation-based $3\mathrm{D}$ Video Prediction4 citations · 2022
- 3LUMOS: Language-Conditioned Imitation Learning with World Models4 citations · 2025
- 4
- 5
- 6Augmenting Action Model Learning by Non-Geometric Features3 citations · 2019
- 7