Reconfiguration for Robot Locomotion

Reconfiguration is generally discussed in terms of task-specific shape transformation, but it can also be used for locomotion. We have developed a distributed locomotion algorithm for unit-compressible robots using inchworm-like motion, and implemented this algorithm in hardware on the Crystal system. We also performed extensive experimentation; the algorithm ran for over 75 hours in total at the SIGGRAPH and AAAI conferences. The algorithm and experiments are described in this section.

Inchworm locomotion uses friction with the ground to move a group of unit-compressible modules forward. The algorithm is based on a set of rules that test the module’s relative geometry and generate expansions and contractions as well as messages that modules send to their neighbors. When a module receives a message from a neighbor indicating a change of state, it tests the neighborhood against all the rules, and if any rule applies, executes the commands associated with the rule. The algorithm is designed to mimic inchworm-like locomotion: compressions are created and propagated from the back of the group to the front, producing overall motion.

The message types it can send and receive, and the procedures that are called from the message handlers (including the rules of the algorithm). The “tail” module contracts first, which signals its forward neighbor to contract. Each module expands after contraction, so that the contraction propagates through the robot. When the contraction has reached the front of the group, the group will have moved half a unit forward (in theory; empirical results show nearly optimal distance-per-step for chains of five or more units.

Depending on context, once the leader of the group has contracted and expanded, it can then send a message back to the tail to initiate another step. We implemented this algorithm and performed experiments with various shapes. The experiments successfully demonstrated reliable locomotion in the configurations we tested. See Butler, Fitch and Rus for further discussion. This locomotion gait is significant first in that it exemplifies the style of distributed, scalable algorithms we wish to develop and implement in proposed work.