Robots are increasingly expected to operate in environments that are dynamic and shared with humans. Our research focuses on enabling robots to perceive such environments, learn new skills and interact safely when physical contact is involved. 

A central theme in our work is the combination of vision and touch. While visual perception provides global scene understanding, it may fall short in close-contact situations. Tactile sensing and haptic feedback can fill this gap, allowing robots to grasp objects reliably, manipulate deformable materials and perform tasks where physical interaction is essential. 

Research in this area is key to a wide range of applications: from teleoperation and dexterous manipulation to assistive and medical robotics (particularly in elderly care), robotics for laboratory automation and life sciences, along with humanoid and legged systems, mobile robots, and more general forms of robot-supported human interaction. Across these domains, our goal is to make robots more capable, safer, more intuitive to control, and easier to teach. 

Key Topics

• Robot Learning from Demonstration and Reinforcement Learning
• Robot Foundation Models
• Perception for Robotic Autonomy
• Human–robot Collaboration and Interaction 

Our Work

Our research aims to enable robots to learn directly from human demonstrations through simulation, kinesthetic teaching, teleoperation, and interactive modalities. We investigate how robotic skills can be acquiredthrough reinforcement learning, adapted online to changing environments, generalized to new tasks, and transferred reliably to real-world systems. This integrates robot learning, sim-to-real transfer, motion planning, and control, with particular attention to robustness and operation under constraints. 

We utilize multimodal foundation models for perception and action and investigate how these models can be extended to incorporate additional modalities (vision, language, touch, and radar) while supporting parameter-efficient training and adaptation. We also study architectural improvements that enhance world understanding, long-horizon reasoning, and asynchronous decision-making. Through this, we aim to develop general, efficient, and robust embodied intelligence systems. 

Robust perception is the foundation of autonomous behavior. Our research covers visual-inertial SLAM, dense 3D reconstruction, scene completion, semantic scene understanding, LiDAR-based sensing as well as radar-based perception for environments where vision is unreliable - to develop robust perception algorithms. By integrating visual, inertial, and radar sensing, we enable robots to estimate their ego-pose while buildingconsistent representations of their surroundings, even under occlusions, noise, and dynamic changes. The goal is to support robots and autonomous vehicles with stable environmental information for downstream tasks such as navigation, decision-making, digital twinning, and safe operation. 

We also investigate how robots can work alongside humans in a natural and supportive way. This includes shared control strategies, proactive assistance, and interaction models that incorporate human intent, such as hand-motion-based task specification. More broadly, we study human–robot interaction to design systems that are intuitive, predictable, and responsive in collaborative, contact-rich settings. 

Contact

If you are interested in Robot Perception and Robot Learning or would like to learn more about our work, feel free to reach out to Valdrin Aslani, M.Sc. You are also welcome to get in touch directly with the researchers working in this field:

• Valdrin Aslani, M.Sc.
• Cem Eteke, M.Sc.
• Emre Faik Gökce, M.Sc.
• Mahsa Heydari, M.Sc.
• Xin Su, M.Sc.
• Zican Wang, M.Sc.
• Wenxuan Wei, M.Sc.
• Mengchen Xiong, M.Sc.
• Dong Yang, M.Sc.