Reality and Perceptual Gaps in Expressive Swarm Robotics

Ali Imran  ·  Abdalwhab Bakheet Mohamed Abdalwhab  ·  Rafael Gomes Braga  ·  Hélène Duval  ·  Audrey Rochette  ·  Vivek Shankar Varadharajan  ·  Yann Bouteiller  ·  Jean Mazerolle  ·  Giovanni Beltrame ·  David St-Onge

Lab INIT Robots - École de technologie supérieure (ÉTS), Montréal

Part of the DESSAIM project

Overview

Two gaps between a swarm and its audience

Swarm robotics faces a demanding reality gap: discrepancies in sensing, timing, communication, and actuation propagate through local interactions and reshape collective dynamics. Deploying robot swarms in front of live audiences revealed a second challenge - a perceptual gap between the collective organization a swarm physically realizes and the one an observer reads from its motion.

We examine these stages using tabletop, aerial, and human-scale mobile-manipulator swarms executing the same behavioural code in public performances. Cross-platform measurements show that broad behavioural identity survived deployment, while observer-accessible properties such as tempo and directional sharpness varied substantially across embodiments. Audience responses show that meaning attribution and affect depend on the robotic system and need not vary together.

The central argument is not that either gap has been closed, but that human-facing swarm robotics must make both observable and actionable.

The chain

From specified organization to human interpretation

A designer specifies a collective organization; a particular set of bodies realizes it under their own dynamics, sensing, and update rates; an observer then reads an organization from what those bodies do. The two gaps do not add - they compose: the realization step fixes the stimulus that the reading step operates on.

From specified collective organization to human interpretation: the sequential reality and perceptual gaps.
From specified collective organization to human interpretation. The physically realized collective forms the stimulus available to human perception.
Step 1 · assessed from the robots

Reality gap

From the specified collective organization to the collective physically realized on hardware. Discrepancies in sensing, localization, communication timing, or update frequency can propagate through the network and alter group-level dynamics - even when the same decentralized program runs correctly on every agent.

Step 2 · assessed from people

Perceptual gap

From the realized collective to the organization an observer infers. In a decentralized swarm, the organization of interest often lives in the relations among agents - density, alignment, synchrony, and transitions between behavioural regimes - shaped further by scale, morphology, sound, and spatial framing.

The swarms

Three embodiments, one behavioural code

DESSAIM brought together three swarms spanning almost two orders of magnitude in physical scale. Across all of them, the same Buzz behavioural code was deployed through PyBuzz, a software stack that confines platform-specific sensing and actuation to adapter layers - making genuine cross-platform comparison possible.

Sushis tabletop swarm robots

Sushis

Tabletop

Palm-sized differential-drive robots for dense, fast-iteration collective behaviours on a table.

CrazyCognies aerial swarm robots

CrazyCognies

Aerial

A drone swarm bringing collective behaviours into the air.

Doodies human-scale swarm robots

Doodies

Human-scale

Human-scale mobile manipulators that share physical space directly with the audience.

PyBuzz architecture used to preserve a common Buzz behavioural representation across simulation and heterogeneous physical platforms.
PyBuzz architecture used to preserve a common Buzz behavioural representation across simulation and heterogeneous physical platforms. Buzz is a specialized programming language for managing heterogeneous robots swarms, While ROSBuzz ROSBuzz encompasses the Buzz Virtual Machine (BVM) and interface it with the ROS ecosystem.

Full Code    Buzz    ROSBuzz

Beyond the paper

Performances & residences

Over two years, our heterogeneous swarms performed in public across changing venues, different lighting and acoustics, and audiences in close proximity - an ecological testbed no laboratory replicates easily. Below are video clips from the shows and residences that cannot be included in the paper.

DESSAIM - Highlight Video A short montage across the three swarms and multiple venues.
Sushis in action Swarm choreography with tabletop robots
CrazyCognies in action Swarm choreography for CrazyCognies flying robots
Doodies in action Swarm choreography with Doodies human-scale robots
What we found

Key takeaways

01

Organization transfers, expression does not. Behaviours kept their defining collective form on all three bodies, yet their realized timing and directional sharpness differed substantially.

02

Abstraction has limits. Buzz/PyBuzz standardize sensing and motion, but not the physical timescale: the same control-step parameter meant real durations differing by roughly 3.5× across platforms.

03

The collective is the perceptual object. Meaning was read from relations among agents - density, alignment, synchrony, and transitions - rather than from any single robot's trajectory.

04

Embodiment shapes the reading. Sushis evoked collective and biological metaphors; Doodies supported both individual and group-level readings, from graceful to confrontational.

05

Meaning ≠ affect. Greater meaning attribution did not imply more positive affect; legibility, comfort, and acceptance stay distinct dimensions.

06

Performance is an ecological stress test. Public shows couple both gaps at once - technical realization and human interpretation must hold together in the same uncontrolled setting.

Cite this work

Citation

If you find this work useful, please cite it.

@article{dessaim2026, title={Reality and perceptual gaps in expressive swarm robotics}, author={Imran, Ali and Abdalwhab, Abdalwhab Bakheet Mohammed and Braga, Rafael Gomes and Duval, H{\'e}l{\`e}ne and Rochette, Audrey and Varadharajan, Vivek Shankar and Bouteiller, Yann and Mazerolle, Jean and Beltrame, Giovanni and St-Onge, David}, journal={arXiv preprint arXiv:2611.05244v2}, year={2026}, month={nov}}