swarm-robotics

In this talk, we present new results for the algorithm for swarm aggregation by Gauci et al. (2014). We first identify deadlock configurations for $n > 3$ robots that never aggregate. We then show the algorithm’s robustness to noise and error. Finally, we demonstrate how generalizing the algorithm with cone-of-sight sensors can improve runtime.

We answer negatively the open question of whether the swarm robotics aggregation algorithm by Gauci et al. (2014) will always aggregate systems of $n > 2$ robots when the robots' movement is deterministic and uniform. In simulations, we demonstrate that the algorithm is robust to sensor and motion noise. Lastly, we prove that cone-of-sight sensors can provide linear speedup over line-of-sight sensors in the $n = 2$ case.

How can an observed or desired collective behavior be reverse-engineered into local rules that individuals can embody? In this talk, I present a framework for designing stochastic distributed algorithms whose long-run collective behaviors can be formally characterized by phase transitions and can be embodied in analog collectives.

Harnessing phase changes from statistical physics in algorithm design. Incorporating algorithmic theory in robot design and manufacturing. Making analogies between macro-scale robot swarm dynamics and the physics of granular active matter systems.

We study the collective aggregation and dispersion of BOBbots. We demonstrate that a rigorously analyzed algorithm can be translated to these simple robots by mapping probabilities onto the physics of interactions, yielding the desired behaviors without digital computation, communication, or sensing.

In this dissertation defense, I treat three questions on emergent behavior in programmable matter: What are the minimum individual capabilities necessary to achieve a collective behavior? How can existing algorithms be enhanced to capture realistic assumptions? How can digital algorithms be translated for simple, analog systems?

The world is filled with systems of entities that collaborate in motion, both natural and engineered, that are capable of sophisticated emergent behavior arising from simple interactions of their members. This PhD dissertation collects my work on the amoebot model, stateful and stochastic distributed algorithms for programmable matter, and translations to active granular matter.

In this talk, I give a comprehensive overview of the amoebot model for programmable matter, dividing algorithms and applications into three main categories: deterministic/stateful algorithms, fully stochastic algorithms, and applications to swarm robotics.

We study the ensemble behavior of smarticles, smart active particles that are incapable of moving alone, but collectively can push against one another. We show that by introducing asymmetries among smarticles' behaviors, their chaotic interactions can yield directed motion.