It was my expectation that dimensions of the basis could be scaled to reduce force effects along those directions, but the final projection in the force calculation actually requires that dimensions be zeroed. It's only the directions that matter and not the scales. Rather than a simple projection, this should be treated like inference with a multivariate Gaussian. The attractor is then an ellipsoid.

Write some tests for this package. Some options:

• Tests based on the Force Dimension example code.
• Tests for ensuring an adequate update rate.
• Tests that drive the robot to apply impulses and measure the response.
• Tests of the basic math.