RI Seminar: Sam Burden : Toward telelocomotion: human sensorimotor control of contact-rich robot...
Sam Burden, a University of Washington faculty member, discusses how to help people control contact-rich robots, presenting research at a CMU Robotics Institute seminar hosted by Aaron and Matt Travers.
In this talk
- 1:09
Why put people in the control loop
Burden points to surgery, brain-machine interfaces, and a passive walking device that reduced metabolic cost by about 7% as examples of machines assisting or enhancing human ability. He focuses on teleoperation for dangerous, dynamic work such as firefighting, where people could guide robots while autonomy remains limited. He cautions that human sensorimotor delay can destabilize control, as studied in aerospace, and that safety features can change how people respond.
- 9:07
Modeling contact-rich robot dynamics
Burden describes legged robots as hybrid systems: their dynamics change abruptly as limbs make or break contact, and impacts reset velocity. With rigidly coupled limbs, small changes in initial conditions can produce discontinuous differences in trajectory outcomes, complicating planning and optimization. In the mathematical models he presents, decoupling limbs can restore continuity, while a viscoelastic spring-damper can support contraction by smoothing force changes at impact.
- 21:41
Contraction and its design tradeoffs
Burden defines a distance that treats pre-impact and post-impact states at the same instant as zero distance, then combines continuous dynamics with impact behavior to analyze contraction. Under the stated conditions, trajectories converge exponentially toward one another, including toward time-varying behaviors driven by an input. The required limb decoupling may add degrees of freedom and nonlinearities, and the theory has an obstacle around grazing contact.
- 39:46
How people learn to control systems
Burden studies human operators using linear systems as a tractable starting point for understanding control and learning. He says people use both feedforward information from a reference and feedback from tracking error, and learn to approximately invert a system's dynamics. In experiments, participants using surface EMG had lower inversion error for second-order systems than those using a manual interface, while tracking error was similar.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.