Behaviors for Compliant Robots

1
Behaviors for Compliant Robots Benjamin Stephens
Christopher Atkeson
Experimental Setup and Results
Full Body Control and Estimation We implement
full-body force control on the humanoid robot.
The combination of the high degree-of-freedom and
force control make the system suitable for
complex interaction tasks such as helping a
person lift a table (left). To perform such
tasks, we implement disturbance-modeling state
estimation (right)

• Hardware Platform
• We implement our work on the Sarcos Primus
  humanoid robot. It has several unique features
• High degree of freedom
• Full body force control
• Located in a motion capture laboratory
• Capable of human-like forces and speeds

Leveraging Human Motion Capture Another technique
we are using is taking human motion capture to
design full body trajectories. This simplifies
motion design which would otherwise be very
tedious if hand-tuned. Because the robot is
dynamically different from the human, we use the
motion capture as a reference which the full body
controller tries to imitate while coordinating
other objectives such as balance and effort. We
have implemented standing dance motions on the
robot (right).
Simple Models for Push Recovery Full body control
is driven by simple models of the center of mass,
angular momentum and footstep locations.
Controllers are designed for these simplified
models using advanced techniques such as optimal
control. Our primary task is push recovery, for
which we have demonstrated multiple strategies
(right) ankle strategy, hip strategy and
stepping.