Inspired by collectives in nature,
robots may be guided by decentralised
algorithms, independent of key
individuals and single points of
failure. Such systems rely on simple
robots that may realistically and
reliably perform the tasks they are
given: devoid of global awareness
that requires expensive sensors; of
global communication that may result
in bandwidth issues with increasing
populations; and of complicated
mechanical manoeuvres that are likely
to have low success rates. Biologically
inspired control methods dependent
on locally propagated gradients,
positive and negative feedback
mechanisms, stochasticity, stigmergy
and environmental templates may
be employed. Evaluation criteria
include power and time efficiency,
convergence guarantees, scalability,
and robustness to individual failures.
The projects outlined below focus on
two types of systems: the first leads to
deterministic structures, the second to
approximate structures according to
environmental templates or functional
goals. Although the methodology of
collective embodied intelligence relies
on simple individuals, it can result in
robust and complex behaviours.
COLLECTIVE CONSTRUCTION
IN RESEARCH
The Termes project,1 started in 2009 at
the Harvard School of Engineering and ...
