Mateusz Jaszczuk, a master’s candidate in Mechanical Engineering and Applied Mechanics (MEAM), has completed his thesis, “Learning to Feel: Force-Aware Data-Driven Estimation and Control for Adaptive Physical Interactions,” under the supervision of Nadia Figueroa, Shalini and Rajeev Misra Presidential Assistant Professor in MEAM. His work advances how torque-controlled robots safely and effectively interact with humans in dynamic, uncertain environments.
As robots increasingly operate in human-centered spaces, their ability to respond safely and adaptively to physical contact is critical. Impedance control enables robots to remain compliant during interactions, but its performance depends on accurate models of both the robot and its environment. Mismatches in these models can lead to unsafe behavior or reduced task accuracy.
In his thesis, Jaszczuk develops an adaptive impedance control framework that improves both safety and performance. The approach introduces two key capabilities:
- Online system identification, which enables real-time estimation of unknown loads, such as attached tools
- An interaction classifier that distinguishes between payload changes and external perturbations, such as human contact
Together, these contributions allow robots to adapt to changing conditions while maintaining stability and precision. The system also enables robots to interpret human-applied forces as guidance, supporting more intuitive human-robot interaction.
Jaszczuk validated the framework through experiments in which a human operator attaches unknown objects and physically guides the robot toward a target. The system demonstrated rapid adaptation to changing dynamics while maintaining safe and responsive behavior.
A particularly memorable milestone came when extending the framework to a bimanual manipulation setup. “Coordinating two arms while maintaining adaptive control behavior introduced additional complexity, and getting the system to perform reliably was both challenging and rewarding,” Jaszczuk said. This extension highlights the framework’s potential for more complex, coordinated interaction tasks.
Reflecting on his decision to pursue a thesis, Jaszczuk emphasized the value of working on an open-ended problem. “I wanted to experience the full research pipeline, from identifying a limitation in existing impedance controllers to designing a framework and implementing it on real robotic hardware,” he said.
Jaszczuk’s work contributes to the development of robotic systems that are not only precise but also responsive to physical interaction, an important step toward safer and more effective human-robot collaboration.