Joint Control of Digital Swarms

Particle bodies make control a boundary question

A shared swarm body is both an interface and a way of staging the relation among people.

The broader Plasmatic Multitudes argument begins with bodies whose edges can soften while their coherence survives. David Glowacki’s Isness work gives this question an explicitly social form. Participants appear as luminous energetic essences with diffuse boundaries in a shared virtual environment designed to loosen conventional self–other separation. The empirical study reports selected self-report similarities with psychedelic phenomenology, including connectedness and ego attenuation, but it does not establish that VR reproduces psychedelic states as a whole (Glowacki et al., 2022).

Glowacki’s later account describes a wider design space of weakly representational, light-energy bodies. Low structural specificity and low symbolic rigidity leave room for participants to project meaning into an intentionally under-specified form. In that setting, a boundary is not merely the outline of an avatar. It regulates how strongly a body reads as separate, how easily it can coalesce with another, and how much interpretation the representation invites (Glowacki, 2024).

John Desnoyers-Stewart’s sequence makes the particle-body side of this problem more operational. In Body RemiXer, tracked people appear as particle clouds and three-dimensional silhouettes. Touch connects their auras through a steady particle exchange, and body swapping superimposes one person’s aura onto another. In Star-Stuff, two participants meet as galaxies and constellations rather than humanoid avatars. Embodied Telepresent Connection uses particle-based aura avatars, cloud-like hearts, sound, simulated physics, and biosignal-linked feedback to support remote pseudohaptic touch (Desnoyers-Stewart et al., 2020; Desnoyers-Stewart, 2022; Desnoyers-Stewart et al., 2023).

These works show why shared control cannot be treated only as command allocation. When a body carries intimacy, coalescence, or self–other ambiguity, an authority policy also shapes the experienced boundary. A participant may feel that a movement is mine, ours, negotiated, resisted, or produced by the system. The visual form alone cannot explain that difference. The system must expose who acted, what was accepted, what was transformed, and what the collective process did on its own.

A digital swarm is a medium, not an unfinished robot

Physical robots expose material constraints. Digital swarms expose a different design space.

Particle systems made birth, change, and death ordinary operations of computational matter. Flocking models showed how local rules can produce coherent aggregate motion without scripting every trajectory (Reeves, 1983; Reynolds, 1987). A digitally native swarm can extend those foundations without inheriting wheel dynamics, drone flight, battery limits, or a requirement to transfer to hardware.

Its native operations may include birth, death, duplication, replacement, merge, split, resampling, graph rewiring, procedural growth, or conversion among particles, clusters, surfaces, and fields. It can move through non-Euclidean space, cross through itself, reverse time, or preserve incompatible outcomes as separate branches. Those operations are not convenient approximations of robot behavior. They define the ontology of the digital body.

Robot swarms remain useful comparison systems because they force questions about collision, connectivity, communication loss, energy, maintenance, and physical safety. They are a later translation branch, not a maturity ceiling. A rigorous digital experiment should be judged by whether its participants, interaction, technical behavior, and interpretive claims match the digital substrate it addresses.

A single session may pass among all four. Changing scale therefore changes more than the camera zoom. It changes what kind of thing a person is controlling and what can count as continuity.

Shared control is a stack of decisions

“Collaborative,” “shared,” and “multi-user” remain vague until the system states who can affect what and what happens when intentions overlap.

Robot-swarm research offers a useful command vocabulary. Kolling and colleagues distinguish behavior or algorithm selection, parameter setting, environmental influence, and influence through selected members or leaders (Kolling et al., 2016). These categories describe ways human intent can reach a distributed process. They do not by themselves describe multi-user authority, morphology, representation change, or collaborative history.

A broader digital system needs at least five separate decisions. Binding determines which people can affect which parts, regions, scales, roles, or functions. The control target specifies what they may change. Arbitration resolves simultaneous or conflicting intentions. Execution determines how local rules, constraints, or learned systems interpret an accepted command. Feedback and history show what happened, why it happened, and whether it can be reversed.

What a participant may influence

These targets can coexist. One person may shape a movement field while another edits density and a third controls when a proposed merge becomes permissible. Different inputs can also reach the same semantic action. A switch, a hand gesture, a line of text, or a pre-authored pattern can all issue a consequential command without pretending that the bodily acts are equivalent.

How several intentions become one event

Arbitration is not an implementation detail. It determines whether a shared body behaves like a collection of personal territories, a negotiated instrument, a majority system, an unstable compromise, or a field that preserves several forces at once.

A seventh possibility is to retain conflict as visible motion: competing attractors, oscillation, seams, or local instability. Mass collaborative systems show that minority action, overwrite, competition, and repair can materially shape a collective result. They do not prove that participants will experience real-time swarm conflict as expressive rather than frustrating. That remains a design hypothesis (Aleta and Moreno, 2019; Müller and Winters, 2018).

Mathematical influence is also not the same as felt agency. In virtual co-embodiment, paired participants shared one avatar through weighted averages. Participants could estimate their objective contribution, yet predictable movement could produce more agency than the control weight alone would suggest. A later study found that haptic feedback and weight distribution interacted with the task rather than improving agency in one simple direction (Fribourg et al., 2021; Venkatraj et al., 2024).

For a plasmatic body, the consequence is immediate: contribution weight, visible causal effect, perceived agency, bodily ownership, and authorship must be measured separately. A person may shape a transition without owning the resulting body, feel agency with little objective influence, or regard a sparse but decisive veto as their most meaningful contribution.

Existing systems establish fragments of the design space

The strongest evidence comes from several neighboring fields. Their boundaries matter as much as their contributions.

What the robot-swarm comparison contributes

Multi-operator robot research makes authority transfer and resource sharing explicit. In Miyauchi, Lopes, and Groß’s paired study, operators normally controlled dedicated leader-centered teams and could send or request followers while the controller preserved connectivity. Robot sharing did not produce a simple task-performance win. Direct communication improved task score; sharing primarily reduced movement and energy and was perceived as useful (Miyauchi et al., 2023).

The useful transfer is organizational: fixed allocation, shared pools, temporary exchange, functional roles, overlapping scopes, hierarchy, and adaptive allocation are different ways of binding people to a population. The robots also retain local safety and connectivity constraints even when people redistribute authority. A digital body can borrow those distinctions without pretending that physical connectivity or robot ownership defines its expressive possibilities.

A comparative programme for one shared body

The first useful experiment is a comparison of authority policies, not a search for one supposedly natural interface.

Two or three participants could guide one field-driven particle body through a short sequence: gather, divide, hold, yield, merge, recover, and re-individuate. The task should remain simple enough that a failure can be traced to a control policy, a mapping, a rendering, the autonomous solver, or social coordination.

Measures must follow the event

Technical measures include input-to-acceptance latency, input-to-visible-effect latency, stability, command rejection, conflict frequency, participation concentration, solver cost, recovery time, replay accuracy, and morphological continuity. Task performance is useful, but it cannot establish meaningful participation or shared authorship by itself.

Participant measures should separate agency, ownership, authorship, causal understanding, fairness, expressive range, collaboration, access cost, and willingness to preserve the result. Workspace-awareness research supports showing who is acting, where, on what, and what they are likely to do next (Gutwin and Greenberg, 2002). The Virtual Embodiment Questionnaire and Creativity Support Index offer useful components, but neither can be transplanted without adaptation: one assumes a virtual body, while the other does not measure causal authorship or authority (Roth and Latoschik, 2020; Cherry and Latulipe, 2014).

Mixed-ability participation makes this comparison more exact rather than peripheral. High-rate motion, speech, gaze, touch, switches, text, facilitated input, and pre-authored patterns should be able to reach meaningful semantic actions. A slower or intermittent contribution must not become less authoritative merely because it produces fewer events. The dedicated Mixed-Ability Human-Swarm Interaction page develops the access, mapping, consent, privacy, refusal, and repair requirements for that work.

An architecture for attributable transformation

The human-facing system should send semantic intentions rather than raw controller events to the swarm. An intention records the participant, action, target, scope, desired timing, strength, duration, branch, uncertainty, and attribution policy. The authority layer then decides whether the action is accepted, transformed, delayed, combined, refused, or branched.

The execution layer remains free to use particles, boids, fields, graph relations, state machines, procedural morphology, or learned policies. It must return more than the final animation. Participants need to know which elements were affected, which constraints changed the command, whether another intention interfered, and whether the result is continuing, complete, or failed.

This separation also makes undo and replay more honest. A collaborative undo should add a compensating historical event rather than silently erase evidence that an action occurred. Incompatible transformations may need branches instead of a forced merge. Periodic snapshots, semantic events, solver version, random seed, and dependency information make it possible to reconstruct how a form arose.

Tölvera is a plausible open execution substrate for flocking, swarming, artificial life, OSC, and interactive mappings, though it has no native multi-user authority model. HuGoS provides a browser-delivered many-agent experiment pattern. Estuary demonstrates a shared generative process with different participant views. None is a turnkey solution; together they suggest a modular prototype with a semantic server, interchangeable arbitration, a particle or boid adapter, participant clients, and an append-only event log (Armitage et al., 2024; Ogborn et al., 2017).

Physical embodiment is a translation problem

A later physical system does not have to reproduce every virtual particle. A small robot group might render landmarks, subgroup boundaries, direction, tension, or warnings while the digital body remains much larger. The semantic command and event history can persist while the embodiment changes.

Translation must preserve authority, consent, attribution, and repair rights as well as movement. Physical robots add hazards and maintenance responsibilities that a reversible particle system does not have. Conversely, a robot test cannot validate instant topology change, branching history, interpenetration, or field-like embodiment when those are the research object.

This is why physical validation remains one branch of the programme. It tests which semantic operations survive material constraints. It does not decide whether a digitally native collective body is legitimate.

What the evidence supports

The central research object is therefore not a more efficient command panel. It is a body whose boundaries, internal organization, authority, and history can all become shared material. The body may transform the participants’ intentions as it executes them, but it should not hide that transformation.