Bugs with Features: Vision-Based Fault-Tolerant Collective Motion Inspired by Nature

TL;DR

The paper tackles brittleness in vision-based swarm robotics caused by perceptual ambiguities and occlusions. It introduces two locust-inspired mechanisms: a robust distance estimator that fuses vertical and horizontal visual cues ($r_L$, $r_R$, and corrected center distance $r_C$) and an intermittent Pause-and-Go locomotion strategy that enables nominal robots to observe and identify faulty peers, mitigating their disruptive effects. The authors formalize an AA-V model for vision-based sensing and an AAPG-V extension that partitions neighbors into nominal and faulty sets, using stochastic handling to manage misclassifications. Through extensive physics-based simulations in ARGoS3, the approach yields substantial improvements in collective-order robustness and resilience to faults across various swarm sizes and models, with practical implications for real-world vision-enabled swarms.

Abstract

In collective motion, perceptually-limited individuals move in an ordered manner, without centralized control. The perception of each individual is highly localized, as is its ability to interact with others. While natural collective motion is robust, most artificial swarms are brittle. This particularly occurs when vision is used as the sensing modality, due to ambiguities and information-loss inherent in visual perception. This paper presents mechanisms for robust collective motion inspired by studies of locusts. First, we develop a robust distance estimation method that combines visually perceived horizontal and vertical sizes of neighbors. Second, we introduce intermittent locomotion as a mechanism that allows robots to reliably detect peers that fail to keep up, and disrupt the motion of the swarm. We show how such faulty robots can be avoided in a manner that is robust to errors in classifying them as faulty. Through extensive physics-based simulation experiments, we show dramatic improvements to swarm resilience when using these techniques. We show these are relevant to both distance-based Avoid-Attract models, as well as to models relying on Alignment, in a wide range of experiment settings.

Figures (20)

• Figure 1: Focal robot’s coordinate system and relative neighbor location. The dashed arc marks the bearing angle $\beta_{ij}$. The thick solid black arrow measures the distance $r_{ij}$. The vertical $x$ axis marks the focal robot's current heading.
• Figure 2: Screenshot from the ARGoS3 pinciroli12 showing a swarm of 40 simulated Nymbots. The gray circle represents the sensing radius ($r_{sense}$) of a robot.
• Figure 3: Evolving order parameter of the system with different swarm sizes.
• Figure 4: Performance comparison of occlusion handling methods across varying $r_d$ and fixed $r_{sense}=0.19\unit{\metre}$. Experiments were carried out in a swarm of 40 robots, using the AA model. Means displayed are across 50 trials per setting. Occlusion-aware models exhibit reduced sensitivity to $r_d$ variations, maintaining more stable collective behavior.
• Figure 5: Neighbor views from different observing positions. (a) 3D rays to the outermost horizontal and vertical edges. (b)--(d) 2D top-down views of the horizontal effective edges, with notation as in \ref{['fig:topdown-length']}.

Theorems & Definitions (1)

• Definition 1