SCIoI-supervised Master thesis receives Rolf-Niedermeier-Award

By: Maria Ott

Adrian Pfisterer, a recently graduated master’s student at Technische Universität Berlin, has been awarded the Rolf-Niedermeier Prize for his innovative thesis addressing challenges in robotic manipulation. His research was supervised by Xing Li and Vito Mengers, and Oliver Brock as advisor, all members of Science of Intelligence (SCIoI). The Rolf-Niedermeier Prize, awarded annually for outstanding academic theses, recognizes exceptional achievements in computer science and related fields. Adrians work, conducted at SCIoI, stands out for its combination of theoretical depth and practical application, demonstrating how robots can better process incomplete or uncertain information to perform complex tasks.

From Human Hands to Robotic Precision

Robotic manipulation of small objects is a notoriously difficult problem. Visual uncertainties, such as occlusions or noisy sensor data, make it challenging for robots to estimate an object’s movements accurately. Adrian’s thesis addresses this by incorporating human hand movements as a guiding framework. By observing how humans interact with objects and explicitly modeling uncertainties in visual data, his method allows robots to estimate an object’s kinematic model in real-time.

“This approach enables robots to handle tasks that would otherwise be infeasible due to the limited accuracy of traditional methods,” explained Vito. Vito’s work focuses on interconnected recursive estimation, which Adrian leveraged to enhance the robots’ ability to learn from human demonstrations.

Collaborative Research at SCIoI

Adrian’s research combines concepts from two ongoing projects at SCIoI. Xing’s project, “Learning to Manipulate from Demonstration,” explores how robots can learn complex tasks by observing human demonstrations, while the project “Differentiable interconnected recursive estimation,” led by Vito, investigates principles of intelligence. Together, these projects provided the foundation for Adrians’s thesis, which tested and validated these methods in a controlled environment.

“This is a great example of interdisciplinary collaboration,” said Xing, who focuses on enabling robots to perform contact-rich manipulation tasks, such as unlocking mechanisms. “Adrian’s work ties together ideas from different areas to solve a very specific and practical problem.”

Oliver Brock, the official advisor of Adrian’s thesis and the principal investigator of both projects, provided critical mentorship and direction throughout the research process. As the spokesperson of SCIoI, Oliver brings extensive experience in building real-world robotic systems and is a leader in fostering collaborations across disciplines, including psychology and behavioral biology, to advance the understanding of embodied intelligence. His guidance was instrumental in ensuring the research’s success and its contribution to the field of robotics.

Measurable Impact

The results of Adrian’s experiments are promising. By modeling uncertainties explicitly, his method outperformed two baseline approaches by significant margins, providing more accurate estimates of object motion. This improvement is particularly important for small objects, where minor inaccuracies can lead to substantial errors.

“Our findings show that integrating human demonstration data with robust modeling of uncertainties can make robots more reliable in handling delicate tasks,” Adrian explained. “It’s a step toward creating robots that are better equipped to assist in real-world scenarios.”

Recognizing Excellence

The Rolf-Niedermeier Prize acknowledges the late Rolf Niedermeier, a professor of computer Science at TU Berlin, known for his research in computational complexity theory, especially in parameterized complexity, graph theory, computational social choice, and network analysis. It is awarded based on academic merit and the quality of the thesis. Adrian’s recognition reflects his hard work and the strength of the mentorship provided by Xing and Vito.

“It’s an honor to receive this award,” Adrian said. “I’m grateful for the guidance I received from my supervisors and the collaborative environment at SCIoI, which made this research possible.”

The award was presented at the faculty’s graduation ceremony on December 6, 2024, celebrating academic excellence and innovation. Adrian’s achievement highlights the potential of interdisciplinary research in advancing robotic systems capable of tackling real-world challenges.

Research

An overview of our scientific work

See our Research Projects