Our work at CMU's Robotics Institute working on HRI with CoBots this summer.

Tuesday, July 14, 2015

Laban Effort Test Cases

Space

The following features of space path trajectories will be used for the calibration pilot test.

Path shapes:
  • straight, curved, sinusoidal, exponential sinusoid
Orientation:
  • attentional goal, center of path, avoidance, fixed angle(?)
Distance between waypoints:
  • curvilinear, rectilinear
Amplitude, frequency, exponent, and linearity can be specified along their respective gradients, and should be tuned to produce trajectories that are the best representations of Laban Effort states.

In subsequent testing, other explorable cases can involve:
  • comparing paths in differently sized spaces, i.e. an open area vs a hallway.
  • varying the orientation by means of a Gaussian distribution centered on the calculated orientation

Time

Speed:
  • 0.2 m/s, 0.6 m/s, 1.0 m/s along a fixed distance
Acceleration:
  • constant acceleration and deceleration patterns at +-0.1 m/s^2 (possibly also at +-0.2 m/s^2?)
Hesitation:
  • patterns of hesitation with constant speed before and after hesitation at 0.2 m/s, 0.6 m/s, and 1.0 m/s
  • Should I do slow hesitation still or only stop?
  • Need to be sure that distance is consistent

Friday, July 10, 2015

Laban Space Paths

Path Shape:

Any of these path shapes can be flipped vertically by specifying a negative parameter for the path. 

Normal (y=C)

 The normal path is a straight line from the startpoint to the endpoint. This can be considered the most direct path shape.

Curve (y=sin(0.5x))
 
By raising the exponent, the curve path becomes more gradual near the beginning of the path and sharper at the end of the path.

Curve^2 (y=sin(0.05x^2))

 Curve^3 (y=sin(0.005x^3)
 Curve^4 (y=sin(0.0005x^4)
Curve^5 (y=sin(0.00005x^5)

Orientation along the Path:
  • Look along center angle of the path (determined by slope of the path at each waypoint)
  • Look at attentional goal
  • Switch between looking along center angle of path and looking at attentional goal
  • Keep a constant orientation along the entire path (along x-axis)
Variance of Orientation:

A higher variance will cause the orientation at each waypoint to be sampled from a larger Gaussian distribution around the selected orientation setting. A variance of 0 will have no variance from the selected orientation.
Segmentation:

Paths can also be generated by dividing the distance between the startpoint and the endpoint into different path segments.  

Hook (normal + curve)
Sine wave (curve + curve)

Moving Backwards:

If the back parameter is set to True, then the Cobot will switch directions and move backwards to a prior waypoint before then continuing forward along the path.

Waypoint Separation:

If waypoints are closer together, the path will appear smoother and curvilinear. If waypoints are farther apart, the path will appear more jagged and rectilinear.


Wednesday, July 8, 2015

Initial Time Movement Patterns

Shown below are graphs of initial patterns of movement for exploring the Time state. Acceleration is used in both the positive and negative directions (shown by red and blue lines).

Parameters are given to explain how these patterns are created.
Example command:
python trajectoryWithTime.py [start_x] [start_vel] [acceleration] [num_waypoints] [hesitation]

These graphs show the position, velocity, and acceleration of the constant accelerating and decelerating patterns. Based on testing with the real robot rather than the simulator, I have eliminated all 0.1 m/s velocities because it appears to be below the robot's threshold for movement (to be looked into more thoroughly in the future). This includes the first and last waypoints for acceleration and deceleration, respectively.

Example commands:
acceleration: python trajectoryWithTime -16 0.2 0.1 10 F
deceleration: python trajectoryWithTime -16 1.0 -0.1 10 F





These graphs show the position, velocity, and acceleration of the paths that accelerate or decelerate after a hesitation. Again the waypoints at speeds less than 0.2 m/s were eliminated from the motion patterns. This includes the first and last waypoints for acceleration and deceleration, respectively as well as changing the velocity of the hesitation from it's original value of 0.05 to 0.2 m/s.

Example commands:
acceleration: python trajectoryWithTime -16 0.2 0.1 10 T
deceleration: python trajectoryWithTime -16 1.0 -0.1 10 T

Note: I have been switching between using 0.1 m/s (accel) + 1.0 m/s (decel) and 0.5 m/s for start velocities. In this case, start velocity is the velocity after the hesitation. It travels at 0.5 m/s to the point of hesitation in both conditions. This needs more work... not sure the best way to make these conditions comparable as acceleration and deceleration look very different after a hesitation.