Supplementary Video for the Instructional Improvement Program Grant Application: "Multi-Media Courseware in Robotics"

Prof. Katie Byl, ECE Dept. (

Rovio: a 3-wheeled robot

Below is a short video of the Rovio when commanded to move in a straight line to the left.

The robot has intentionally been aligned with the floorboards at the start, to make it easier to tell if it is traveling in a straight line. Note that it actually turns significantly. In one anticipated Rovio-based projects, students will design feedback control to correct for these errors and track motions desired motion plans much more accurately.

Underactuated dynamic systems

One exciting - and often unavoidable - challenge in robotics in the problem of underactuation: Sometimes, there are more degrees of freedom in a system than we can control directly with the actuators we have. Below are several examples of underactuated systems.

Hovercraft - 3 degrees of freedom, but only 2 actuators.

Below is an animation of a "hovercraft" which has 3 degrees of freedom but only 2 thrusters (the actuators), making it underactuated. The degrees of freedom are: (1) x position, (2) y position, and (3) the rotation angle, theta.

Here, we can control any 2 out of 3 degrees of freedom, but not all 3. Thus, if we want to follow an (x,y) path, the angle theta is left uncontrolled, and the hovercraft spins. The Rovio, by comparison, has the same 3 degrees of freedom (x, y, and theta), but it has 3 actuators (the 3 wheels), making it fully actuated. Unlike the hovercraft below, the Rovio can therefore follow a desired motion path in x, y, and theta.

Acrobot - 2 degrees of freedom, but only 1 actuator (at "elbow").

The acrobot is a two-link pendulum with a single actuator at the elbow that can be (1) brought into an upright position and (2) stabilized. Below are two methods for achieving the first goal of pumping energy into the system to raise both links; an additional, linearized controlled can than be added for stabilization.

The first videos shows colocated partial feedback linearization (PFL), while the second demonstrates non-colocated PFL.

Note that while the second method is much faster at getting upright, it also requires a HUGE amount of power initially to do so! It may work in simulation, but would not work on practical hardware. Also note the long-term behavior of each approach: this method does NOT stabilize the pendulum at the top; it only raises it into a position where a second control algorithm could take over to do so.

Marc Raibert's one-legged hopper.

The final simulation example shows the Raibert hopper. It is impressive that Raibert's simple balance algorithms can stabilize this underactuated system. However, the resulting speed control is not very accurate. Note that Raibert's control ideas, which were first developed in the 1980's, are at the basis of control for the BigDog robot that you may seen on YouTube in recent years.