Aggregating swarms through morphology handling design contingencies: from the sweet spot to a rich expressivity

TL;DR

The study investigates how robot morphology, via self-alignment exoskeletons, interacts with locomotion policies to enable phototactic aggregation in swarms. Combining 64-Kilobot experiments with a faithful self-alignment model and in silico sweeps, it shows that fronters can leverage a motility-induced phase separation–like mechanism under a non-stop slow-down policy, while aligners fail to aggregate; a narrow sweet spot in the parameter $\epsilon/\tau_n$ is required for efficient aggregation, and broadening the morphology–policy space reveals a spectrum of collective behaviors with programmable expressivity. These findings bridge active-matter theory and swarm robotics, highlighting adaptive morphology and morphology–policy co-design as routes to robust and expressive swarms. The work suggests future directions in reconfigurable morphologies and embodied learning to enable swarms that switch among collective modes in context, leveraging physical interactions as implicit computation.

Abstract

Morphological computing, the use of the physical design of a robot to ease the realization of a given task has been proven to be a relevant concept in the context of swarm robotics. Here we demonstrate both experimentally and numerically, that the success of such a strategy may heavily rely on the type of policy adopted by the robots, as well as on the details of the physical design. To do so, we consider a swarm of robots, composed of Kilobots embedded in an exoskeleton, the design of which controls the propensity of the robots to align or anti-align with the direction of the external force they experience. We find experimentally that the contrast that was observed between the two morphologies in the success rate of a simple phototactic task, where the robots were programmed to stop when entering a light region, becomes dramatic, if the robots are not allowed to stop, and can only slow down. Building on a faithful physical model of the self-aligning dynamics of the robots, we perform numerical simulations and demonstrate on one hand that a precise tuning of the self-aligning strength around a sweet spot is required to achieve an efficient phototactic behavior, on the other hand that exploring a range of self-alignment strength allows for a rich expressivity of collective behaviors.

Figures (3)

• Figure 1: Experimental realization of a phototactic aggregation task for $N=64$ robots with morphological actuation through a force re-orientation response. A. Each robot is programmed to perform a Run-And-Tumble motion, with different speeds based on the local light intensity. The run phases in the dark are clearly visible with a greater speed and persistence length. After some time at the boundary, robots eventually get re-injected in the bulk due to the orientational noise. Note that the typical time spent at the boundary is slightly greater for fronters than aligners due to the morphological response ($\approx$ x1.4). B. Effective velocity of robots (aligners) in the dark and light region, measured from the mean square displacement. C. Kilobot rubenstein_kilobot_2012 Robots are embedded in exoskeletons replacing their base legs. The contact point asymmetry and uneven mass distribution lead to a re-orientation response when subjected to external forces. The aligner aligns towards external forces while the fronter aligns in the opposite direction. During a collision, morphological actuation plays a role for robot aggregation : the force re-orientation response of the fronter favors aggregation, provided the intensity of said response is low enough to not cause an unending interlock.
• Figure 2: Experimental results for a phototactic aggregation task of $N=64$ aligners and fronters. A. Snapshot of real robot experiment over time for the two morphologies : aligners on top and fronters on the bottom. The illuminated area (not visible on the images) is surrounded by a red circle. The positive self-alignment of the aligners prevent them from aggregating in the light region, while the fronters succeed in clustering in and around the illuminated region. B. Fraction of the swarm in the light region as a function of time for 4 independent experiments quantifies the lack of aggregation for the aligners, with no significant evolution in time and a mean $N_\circ / N_{tot}$ around $0.16$. C. Conversely, the fronters manage to hold a consistent filling of the light region following and exponential-like convergence to a plateau value around $N_\circ / N_{tot} =0.4$.
• Figure 3: Fraction of the swarm in the light region (red) and total alignment (blue), averaged over the last $100\tau$ of a $1000\tau$ duration experiment, as a function of $\epsilon/ \tau_n$, for the simulated phototactic aggregation task. The horizontal dotted line is the fraction expected from a random exploration of space by non interacting agents, slowing down in the light with a velocity ratio $v_\circ/v_\bullet=1/3$. The insets display simulation snapshots of the final time state of the system at four different values of $\epsilon/\tau_n$, with typical trajectories over the last 10$\tau$.