Institut für
Robotik und Prozessinformatik

Deutsch   English

Zur Zeit leider nur in Englisch vorhanden...

Estimation of load inertial parameters (Ongoing Project)

Project Description

Why estimate the inertial parameters of a load attached to a manipulator?

Many robotics applications require knowledge of specific properties of a robot end-effector or a gripped object, e.g. its mass, shape, or elasticity. Other parameters of a load - like its complete set of inertial properties - have attracted less attention so far. However, many robotics applications may benefit from knowing the inertial parameters of the load that is attached to the robot. The mass, the coordinates of the center of mass (COM) and the inertia matrix of an object may be employed to

  1. eliminate all inertial forces and associated torques acting on a moved robot load, e.g. gravity and centrifugal forces, in order to improve force control in dynamic situations
  2. recognize objects based on their inertial properties (and complementary information like geometric)
  3. estimate the gripping pose of an object, e.g. to calculate a corrective movement in adaptive assembly tasks.

In order to obtain an estimate of the inertial parameters of a load (including its inertia matrix), the object has to be moved, and forces and torques acting on the object have to measured simultaneously. Based on the fundamental Newton and Euler equations, the inertial parameters may be estimated. However, the estimation process also requires the angular velocity, angular acceleration, and linear acceleration of the load apart from forces and torques. A simplified block diagram of the approach sketched below is shown in the following figure.


block_diagram.gif
Block diagram of the approach.

Preprocessing

The manipulator's joint position signals are Kalman filtered to derive angular velocity and angular acceleration of each joint. Subsequently, these signals are used to calculate the angular velocity and angular acceleration of the sensor frame. Angular velocity as well as angular acceleration signals are provided by two sources. So the signal source used for identification can be selected or additional filters can be employed to fuse the angular velocity and angular acceleration signals originating from the sensors mentioned above.

The mounting plate of the JR3 sensor causes disturbance forces and torques due to its mass. In order to reduce these disturbances, the inertial properties of the mounting plate have been identified once and can be used to eliminate these disturbance forces and torques. Additionally, the disturbing forces and torques of a gripper that is mounted onto the sensor can be compensated in an analogous manner.

Parameter Estimation

The parameter estimation step is performed on-line with a rate of up to 1kS/s. In order to reduce bias of the identification results either a variant of the recursive instrumental variables (RIV) method or a recursive total least squares (TLS) method is used. To account for the time varying offsets of the force/torque sensor, either a reset can be performed prior to identification or these offsets can be identified simultaneously to the inertial parameters.

Preliminary results

The results obtained with several test loads show that the identification approach permits the reliable estimation of the mass and the center of mass coordinates of the test loads. The estimates of the elements of the inertia matrix, however, exhibit a median relative error of approx. 10%. The employed test loads and the robot wrist with the JR3 sensor are shown in the figure below.


setup.jpg
Employed sensor and test-loads.


Further information may be found in our Publications. In case of any further questions please contact Daniel Kubus.

It took 0.24s to generate this page.