Cross-Unit Coordination Through Broadcast

Mechanism

The architecture establishes a shared envelope — a declared region of joint state space within which a defined set of actuation units operate under a common governance authority. The envelope encodes the units' permissible kinematic ranges, the geometry of the shared physical workspace, the ordering relations among action classes (which actions must precede which), and a resource budget covering any shared consumable (battery, communication bandwidth, end-effector wear, common manipulator). Each participating unit holds a credential that binds it to the envelope and authorizes it to participate in joint-plan submission.

When a unit proposes an action, it does not commit the action to its actuators directly. The unit constructs a candidate joint plan: its own proposed action, expressed in the envelope's coordinate system, together with a declared time window and a declared resource draw. The candidate is broadcast as a credentialed observation to all peer units within the envelope. Each peer's pre-action gate consumes the candidate together with its own currently-pending or recently-committed actions and evaluates three classes of joint constraint.

The first class is collision geometry: the gate computes the swept volume of the candidate plan over its declared time window and tests for intersection with the swept volumes of peer plans within the same window. The second class is action ordering: the gate consults the envelope's declared precedence relations and tests whether the candidate's start time respects required predecessors and whether its end time permits required successors. The third class is resource budget: the gate sums the declared resource draws of all currently-pending and recently-committed plans, adds the candidate's draw, and tests whether the sum remains within the envelope's declared budget.

A candidate that satisfies all three classes at every peer's gate is instantiated: each peer signs a joint-admission observation, the proposing unit aggregates the signatures into a quorum certificate, and the actuators commit. A candidate that fails at any peer's gate is not instantiated by any unit. The failing peer emits a credentialed non-instantiation observation specifying the failed constraint class; the proposing unit consumes the observation and either revises the candidate or abandons the action. The non-instantiation observation is itself broadcast, so peers that had begun preparatory state changes can roll those preparations back.

The non-instantiation property is structural rather than advisory. A peer's gate that has emitted a rejection observation will refuse to sign a subsequent admission observation for the same candidate without a fresh broadcast carrying a revised plan. The proposing unit's actuator-commit step requires the assembled quorum certificate, so a unit that attempted to commit unilaterally in defiance of a peer rejection would lack the required signatures and would fail its own per-unit governance check. The distributed gate and the per-unit gate together enforce the joint constraint: there is no path by which a non-compliant joint plan can reach the actuators of any participating unit.

Concurrent candidates from multiple proposers are serialized at each peer's gate by a deterministic ordering rule (typically a tuple of the candidate's broadcast timestamp and the proposer's identifier) so that two candidates whose swept volumes intersect are not both admitted. The first-ordered candidate is evaluated against the present pending set; if admitted, it joins the pending set before the second candidate is evaluated; the second candidate's gate evaluation therefore sees the first as already committed and rejects on collision. The ordering rule is itself a declared envelope parameter and is uniform across all peers, so each peer reaches the same serialization decision without coordination.

Operating Parameters

The envelope's collision-geometry constraint is typically expressed as a minimum separation distance plus a swept-volume safety margin; typical values are operation-dependent (centimeters for industrial manipulators, meters for cooperating vehicles, tens of meters for cooperating aircraft). Pre-action gate evaluation latency is typically bounded in the single-digit-millisecond range to single-digit hundreds of milliseconds, depending on the broadcast layer and the number of participating units; the gate is structured so that evaluation completes within the candidate's declared start-window slack. Quorum certificate thresholds are typically set to unanimity among in-envelope peers for safety-critical operations and to a configured fraction (commonly two-thirds or three-quarters) for resource-coordination operations where transient peer non-response should not block progress.

Resource-budget granularity is a declared parameter of the envelope; budgets may be expressed per-time-window (rolling totals over a declared horizon), per-action-class (separate budgets for distinct resource categories), or per-priority-tier (budgets that re-allocate to higher-priority actions when contention occurs). Action-ordering relations are typically encoded as a directed acyclic graph of action classes, with the gate performing a topological consistency check between the candidate and the currently-pending plan set.

Broadcast transport latency is the primary determinant of envelope tightness; envelopes operating over wired industrial networks tolerate finer collision-geometry margins than envelopes operating over wireless mesh links, and the envelope's safety margins are typically configured against the worst-case observed broadcast latency plus a declared safety multiple. Envelope membership may be static (a fixed roster declared at envelope creation) or dynamic (units join and leave under credentialed admission and departure observations). Dynamic envelopes typically require a stabilization window during membership change: a candidate broadcast during the window is held until the new member's gate state is initialized or the departing member's pending plans are drained. Stabilization-window durations are typically configured at single-digit-second granularity and are themselves declared envelope parameters auditable through the credentialed observation stream.

Alternative Embodiments

In a first alternative embodiment, the pre-action gate is centralized: a single envelope-coordinator unit receives all candidate broadcasts, performs the three-class evaluation, and emits joint-admission or non-instantiation observations on behalf of the envelope. This embodiment reduces the per-unit evaluation cost and is appropriate for tightly-clustered fleets with reliable communication to a coordinator.

In a second alternative embodiment, the gate is fully distributed: each unit evaluates the candidate independently against its own state and emits its own admission or rejection observation; the proposing unit aggregates the observations into a quorum. This embodiment is appropriate for fleets without a reliable coordinator and tolerates partial peer failure.

In a third alternative embodiment, the gate operates in two stages: a fast geometric pre-check against simplified swept-volume bounds rejects obviously infeasible candidates within a few milliseconds, and a slower precise check against full kinematic models confirms admission for candidates that pass the pre-check. This embodiment supports high candidate-throughput envelopes such as dense drone formations.

In a fourth alternative embodiment, the envelope supports mixed-autonomy participants: human-operated units broadcast partial-fidelity intent (declared trajectory bounds rather than full plans), and the gate evaluates joint constraints against the partial-fidelity envelope. Autonomous units broadcasting full plans participate alongside human-operated units broadcasting bounds, all within a single coordination framework.

Composition With the Wider Architecture

Cross-unit coordination composes with the per-unit governed-actuation layer that gates each unit's individual actions, with the credentialed-observation broadcast layer that distributes candidate plans and admission observations, with the envelope-credentialing layer that defines who participates in joint-plan submission, and with the rollback layer that retracts preparatory state when a non-instantiation observation arrives. The structural property is that joint coordination is not a separate subsystem layered above per-unit governance; it reuses the same broadcast, admission, and credentialing machinery and adds only the joint-constraint evaluation at the gate.

Composition extends to adversarial-aware operation. Units classified as adversarial or partially-trusted by the envelope's credentialing policy may participate in the gate as observation sources without being granted veto authority; their broadcast plans inform the evaluation but their admission signatures are weighted or excluded according to declared trust policy. The gate therefore operates correctly in mixed-trust environments without requiring a separate trust-management subsystem.

Composition further extends to envelopes that are themselves nested. A higher-level envelope may declare sub-envelopes, each governing a tighter subset of the joint state space, and a unit operating in a sub-envelope evaluates its candidates against both the sub-envelope's local constraints and the parent envelope's broader constraints. Joint-admission observations propagate up the nesting hierarchy as required, so a candidate that requires resources allocated at the parent level cannot be admitted by a sub-envelope quorum alone. Nesting permits operational structures such as squad-within-platoon coordination, cell-within-line manufacturing coordination, or local-airspace-within-regional-airspace traffic coordination to be expressed within a single architectural framework.

Distinction From Prior Art

Conventional multi-agent coordination systems either reconstruct peer state from sensor observation of physical effects (introducing physical-propagation latency that limits coordination tightness) or rely on a centralized planner that allocates actions to units (introducing a single point of failure and a coordination bottleneck). Conventional collision-avoidance systems gate individual unit motion against observed peer motion but do not evaluate joint plans against shared resource budgets or action-ordering constraints. Conventional resource-allocation systems schedule shared resources but do not perform pre-action joint-plan evaluation against geometric collision constraints.

The disclosed architecture combines properties not jointly present in prior art: a pre-action gate evaluating collision, ordering, and resource constraints in a single joint-plan check; non-instantiation as the default for non-compliant joint plans; and the use of the same credentialed-broadcast machinery that handles per-unit governance to handle cross-unit coordination, with no separate coordination subsystem.

Disclosure Scope

This disclosure covers the mechanism by which multiple actuation units coordinate within a shared envelope through a pre-action gate evaluating joint constraints, the use of credentialed broadcast to distribute candidate plans and admission observations, the three-class constraint evaluation at the gate, the non-instantiation property for non-compliant joint plans, the deterministic serialization rule that handles concurrent candidates, the nested-envelope composition property, and the alternative embodiments described above. The disclosure does not claim any specific kinematic model, any specific collision-detection algorithm, or any specific broadcast transport; those elements are implementation choices left to the practitioner. The disclosure applies to actuation domains in which a shared envelope can be expressed and a pre-action gate can be evaluated, including but not limited to ground vehicles, aerial vehicles, marine vehicles, industrial manipulators, surgical robotics, and warehouse logistics; the architectural property is independent of the actuation modality and depends only on the envelope, gate, and broadcast machinery described.