Stanford Robotics

Artificial Intelligence Laboratory - Stanford University

Research on Humanoids

Humanoid robots are supposed to coexist with human and work in our daily environment. They can be used to assist and communicate with us in our homes, offices, public spaces, hospitals, disaster areas, etc. When the robot moves in our environment, safety is one of the most important key problems. In the actual environment, multiple contact and unpredictable contact happen between the robot and the environment or between the robot and human. If these contacts are not modeled in the motion controller of the robot, it causes instability of balance and motion control tasks. In this point of view, compliant and robust contact and motion control is important for humanoid robots. One approach for addressing this problem is to provide torque control. The input torque for the system can be designed to compensate for dynamic effect of the system. Decoupled task dynamics can be applied by the Operational Space Formulation which provides the robot with higher performance in position tracking as well as in compliant motion.

In this project, we are making progress towards the interactive multi-contact and control with the environment or with human, based on the following human-oriented approach.

            (1)  Compliant and passive whole-body control
            (2)  Interactive whole-body control
            (3)  Decoupled task dynamics by the Operational Space Formulation
            (4)  Force control for multi-contact
            (5)  Balance control for multi-contact

This project project collaborates with other project to realize behaviore control, sensing and perception skills.

Compliant and passive whole-body control
Compliant and passive motion control is one of the key technologies for the humanoid robot. In this project, the Torque Transformer is applied to control the existing position controlled system by torque command and to compensate dynamic effect of the system in the motion contrl. Dynamic effect of the robot is modeled by the Operational Space Formulation. The torque command from the Operational Space Formulation is transformed into the joint position and the velocity command through the Torque Transformer. [IROS 2008]


Force Control for Multi-Contact

Contact by the end-effector
Active Observer, AOB was applied to achieve adaptive force control for each translational direction. AOB was designed to handle the uncertainties of contact with the environment using the force sensor at the end-effector. Haptic tele-operation allows an operator to control a slave robot using a master device through which the operator can feel the remote environment of the slave robot. [ISR 2008]

       teleopeforcecontrol   operator1

Contact by the link
Contact forces can be programmed at any point on any link for any direction according to the desired contact situation, and the torques associated with the contact forces are computed through the Contact Jacobian. Dynamic effect of the robot for motion in contact is modeled by the Operational Space Formulation. In this framework, the null space control torque is used for motion control and is formulated so as not to affect the contact forces. [ICRA 2010 WS]


Compliant Upper-Body Control on Stable Balance Controller
One of the merit of using the Torque-Transformer is that current position controller can be also used. In this project, to realize the accurate balance control, and stable lowere-body control, which is related with the hardware, current balance controller was integrated to this system. Compliant upper-body control and stable lower-body control were coordinated together to realize compliant and physical interaction with human. [IROS 2009]

Compliant Balance Control
We have a new approach for compliant balance control.


Shuyun Chung
Taizou Yoshikawa