|SUBJECT:||M.S. Thesis Presentation|
|TIME:||Wednesday, November 21, 2018, 11:30 a.m.|
|PLACE:||MRDC Building, 2405|
|TITLE:||Developing an Anticipatory Controller to Improve Performance of a Snake-like Robot in Unstructured Terrain|
|COMMITTEE:||Dr. Daniel Goldman, Chair (Physics)
Dr. David Hu (ME)
Dr. Frank Hammond (ME)
Limbless robots have the potential to help with many possible applications from search and rescue to surveillance. However, their performance in unstructured environments does not currently match that of living systems. In order to understand how collisions with obstacles effect snake robot locomotion, we studied the interactions between the robot and vertical posts. Under normal open loop control where the robot follows the serpenoid equation, we observe that the collision with the post will often cause the robot to reorient and deviate from its open loop trajectory by an angle which we call the scattering angle. Drawing insights from previous experimental results and simulations, we hypothesize that a model of the interaction between the robot and the posts can be used to implement a simple control scheme to control the orientation of the robot using only minimal onboard sensing. These robot-obstacle collisions are characterized by running many experiments to systematically sample all possible contact conditions between the robot and the post which allow us to model and understand the behavior of the robot at contact. To assist with the collection of this data, an automated gantry system was developed to conduct experiments without any human input. Using this model, we develop an anticipatory control scheme to correct for the scattering that results from the collision with the posts. Contact sensing at the head of the snake measures the location and duration of contact with the pegs. This measurement is used to predict the magnitude and direction of the steering behavior and steer the snake to correct for the scattering. Finally, we experimentally validate this controller for a single post as well as a row of five evenly spaced posts and find that the controller reduces the distribution of scattering angles caused by the post and offers further insight into the nature of the robot-obstacle collision.