Results

A web-based platform was created for crowdsourcing controls for the Softworm moving in a straight line. Users could log into the website with the social media accounts in order to sign up for a time to control the robot. While the users were controlling the Softworm, I collected information on what controls the users were inputing and also tracking the robot as it moved through the arena. This data enabled me to analyze inputs to identify whether or not users converged on a strategy for the fastest controls and also correlate inputs to velocities. It was determined that while users averaged inputs of 1.351 s for t1, 0.554 s for φ, and 1.174 s for t2, these results were biased by the initial inputs of t1=1.500 s, φ=0.100 s, and t2=1.500 s. Users instead tended to prefer SMA1 actuation times of 1-1.3 s for t1 and 0.9-1.2 s t2. The results for matched more closely with the average. It was also determined that the velocity of the Softworm in the x and y directions were an order of magnitude less than expected and further investigation into the cause of that was necessary. However, despite the difference in magnitude the expected inverse relationship between and velocity was observed.

Velocity

Velocities were found from the slope of the linear regressions to the position vs. time plots. The slopes of the regression were matched with inputs. Velocities for the head and and tail were analyzed separately in both the x and y direction. Other factors could include time variations in the SMAs, tether forces, and the initial conditions after reset. The original purpose of matching the control inputs to velocities was to be able to tag the controls as fast, average, or slow for use in a genetic algorithm. However, the velocities obtained through crowdsourcing were much smaller than expected.

User Inputs

When users controlled the Softworm they were able to choose the time that each SMA actuated for (t1 and t2) and the phase gap (φ) in between the actuations. Between November 7, 2013 and March 5, 2014, 920 different inputs were tried, 384 of which were unique. The most common input was the initial input, and it was tried 112 different times, 16% of the time. However, most visitors tried values that no one else had tried before, 273 unique inputs making up 39% of total inputs. Sixty-seven inputs (19%) were tried twice, 23 (10% were tried three times. Seven inputs were tried four times, two five times, five six times, five seven times, one eight times, and one nine times.

Comparisson to Model

Comparing the crowdsourced data to the model developed by Takuya Umedaki, I found my velocities were about an order of magnitude smaller. The reason for this discrepancy is likely due to occlusions, tether forces, or initial reset conditions. To determine the exact cause, the user controls should be run back through the robot control software and new tracking data obtained. Images of the Softworm in the arena should be obtained along with the tracking data in order to determine whether occlusions or other tracking errors were the cause of the discrepancy.

However, despite the differences in magnitude, it was not surprising to find that the fastest velocity correlated with a phase gap of 100 ms nor that phase gap and velocity were inversely related.