Difference: GeneralizedBiped (9 vs. 10)

Contribution: The authors provide an integrated strategy for walking controllers that combines a number of techniques used in virtual control strategy over the past few years. Specifically, they describe 4 components:

• Motion Generator : Joint angles on a character are specified over half a walk cycle using splines (then mirrored for a complete cycle); at runtime, PD-controllers seek to track the specified angles.
• Foot Placement : Robust balance and some degree of velocity control is achieved by foot placement planned using an inverse-pendulum model.
• Velocity Tuner : Using a virtual force derived from the difference between the desired and actual velocity of the character and the Jacobian of the center of mass, a torque is determined that is applied when possible at the stance ankle to move toward the desired velocity.
• Gravity Compensation : Similarly, the Jacobian of the center of mass of each link is used to apply torques that negate the effect of gravity, -mg.

Evaluation: The paper describes experiments in which the controller successfully handles variations in gait parameters, like walking speed, and character proportions and morphology. They also demonstrate the generality of their controller by highlighting several tasks such as picking up objects or carrying crates that are easily incorporated into the controller.

Reproducibility: The authors provide high-level descriptions of each of their system components, but do not describe specific parameters or details. However, the authors do provide source code and executables from the project web page, so the results are indeed testable and reproducible.

Improvements: In general, the paper's writing style is clear and logical. The specific contributions, methodology, and limitations of the control system are laid out in the introduction, which helps provide a roadmap for the rest of the paper. One evalution that may have been helpful is a quantitative comparison of the "naturalness" of their gaits compared to motion capture data, as in "Optimizing Walking Controllers". Do the different gaits of varied characters reflect real-world differences correctly (at least for the physically possible characters)?

-- BenHumberston - 25 Nov 2011