What can robot teams accomplish using only onboard
capabilities and resources?
Mobile autonomy holds great promise for applications ranging
from precision agriculture and self-driving cars to
assistive healthcare and warehouse automation. These
applications and many others all require agents to gather
information using onboard sensors, process it with onboard
computers, share it using only onboard communications
hardware, and then make control decisions. Some approaches
to autonomy implicitly or explicitly require external
resources, such as a lab computer to execute demanding
computations. To enable autonomy beyond controlled laboratory settings,
fundamentally new developments are required to let agents
execute all tasks themselves.
One way of doing so is through control-aware computation and
computation-aware control. Agents have limited onboard
computational power, and this must be accounted for in two
ways. First, agents' controllers can account for the fact
that they are driven by imperfect information and the fact that such
information can enter feedback loops. Second, agents'
computations can be modified to account for the impact that
their imperfections will have upon agents'
controllers. Mobile autonomy presents fundamental challenges
because it tightly couples computation and control, though
this coupling can be accounted for and even exploited to
enable greater degrees of autonomy.
Ongoing work in the CORE Lab swarm robotics
testbed is developing new algorithms and controllers to bring
new levels of autonomy to robotics.
