Page 54: of Marine Technology Magazine (September 2019)
Autonomous Vehicle Operations
Tech Files Robotics con? guration. tion, while optimizing the route for speed. To do so, each CVP
In demonstrations in an MIT pool and in computer simu- precomputes all collision-free regions around the moving lations, groups of linked roboat units rearranged themselves CVP as it rotates and moves away from a stationary one.
from straight lines or squares into other con? gurations, such as After precomputing those collision-free regions, the CVP rectangles and “L” shapes. The experimental transformations then ? nds the shortest trajectory to its ? nal destination, which only took a few minutes. More complex shapeshifts may take still keeps it from hitting the stationary unit. Notably, optimi- longer, depending on the number of moving units — which zation techniques are used to make the whole trajectory-plan- could be dozens — and differences between the two shapes. ning process very ef? cient, with the precomputation taking “We’ve enabled the roboats to now make and break connec- little more than 100 milliseconds to ? nd and re? ne safe paths. tions with other roboats, with hopes of moving activities on Using data from the GPS and IMU, the coordinator then es- the streets of Amsterdam to the water,” says Rus, director of timates its pose and velocity at its center of mass, and wire- the Computer Science and Arti? cial Intelligence Laboratory lessly controls all the propellers of each unit and moves into (CSAIL) and the Andrew and Erna Viterbi Professor of Elec- the target location.
trical Engineering and Computer Science. “A set of boats can In their experiments, the researchers tested three-unit CVPs, come together to form linear shapes as pop-up bridges, if we consisting of one coordinator and two workers, in several need to send materials or people from one side of a canal to different shapeshifting scenarios. Each scenario involved the other. Or, we can create pop-up wider platforms for ? ower one CVP unlatching from the initial shape and moving and or food markets.” relatching to a target spot around a second CVP.
Joining Rus on the paper are: Ratti, director of MIT’s Sense- Three CVPs, for instance, rearranged themselves from a able City Lab, and, also from the lab, ? rst author Banti Ghe- connected straight line — where they were latched together neti, Ryan Kelly, and Drew Meyers, all researchers; postdoc at their sides — into a straight line connected at front and
Shinkyu Park; and research fellow Pietro Leoni. back, as well as an “L.” In computer simulations, up to 12 roboat units rearranged themselves from, say, a rectangle into
Collision-free trajectories a square or from a solid square into a Z-like shape.
For their work, the researchers had to tackle challenges Scaling up with autonomous planning, tracking, and connecting groups Experiments were conducted on quarter-sized roboat units, of roboat units. Giving each unit unique capabilities to, for which measure about 1 meter long and half a meter wide. But instance, locate each other, agree on how to break apart and the researchers believe their trajectory-planning algorithm reform, and then move around freely, would require complex will scale well in controlling full-sized units, which will mea- communication and control techniques that could make move- sure about 4 meters long and 2 meters wide.
ment inef? cient and slow. In about a year, the researchers plan to use the roboats to form
To enable smoother operations, the researchers developed into a dynamic “bridge” across a 60-meter canal between the two types of units: coordinators and workers. One or more NEMO Science Museum in Amsterdam’s city center and an workers connect to one coordinator to form a single entity, area that’s under development. The project, called RoundAr- called a “connected-vessel platform” (CVP). All coordinator ound, will employ roboats to sail in a continuous circle across and worker units have four propellers, a wireless-enabled mi- the canal, picking up and dropping off passengers at docks crocontroller, and several automated latching mechanisms and and stopping or rerouting when they detect anything in the sensing systems that enable them to link together. way. Currently, walking around that waterway takes about 10
Coordinators, however, also come equipped with GPS for minutes, but the bridge can cut that time to around two min- navigation, and an inertial measurement unit (IMU), which utes.
computes localization, pose, and velocity. Workers only have “This will be the world’s ? rst bridge comprised of a ? eet of actuators that help the CVP steer along a path. Each coordina- autonomous boats,” Ratti says. “A regular bridge would be tor is aware of and can wirelessly communicate with all con- super expensive, because you have boats going through, so nected workers. Structures comprise multiple CVPs, and indi- you’d need to have a mechanical bridge that opens up or a vidual CVPs can latch onto one another to form a larger entity. very high bridge.
During shapeshifting, all connected CVPs in a structure But we can connect two sides of canal [by using] autono- compare the geometric differences between its initial shape mous boats that become dynamic, responsive architecture that and new shape. Then, each CVP determines if it stays in the ? oat on the water.” same spot and if it needs to move. Each moving CVP is then To reach that goal, the researchers are further developing the assigned a time to disassemble and a new position in the new roboats to ensure they can safely hold people, and are robust shape. to all weather conditions, such as heavy rain. They’re also
Each CVP uses a custom trajectory-planning technique to making sure the roboats can effectively connect to the sides compute a way to reach its target position without interrup- of the canals, which can vary greatly in structure and design.
September 2019 54
MTR #7 (50-63).indd 54 9/4/2019 9:49:31 AM