Agent-aware coupling
The system is agentic because coupling is decided per agent, per moment, against the applicable profile, task state, and safety envelope. The surface identifies whether the participant is crew footwear or a SLED-class interface, then applies the appropriate coupling regime.
A crew member walking across one tile and a SLED holding position on the adjacent tile can receive different coupling regimes simultaneously (capture-hold-release timing for the crew member’s gait, steady hold-down for the SLED), each tuned to its own profile. Selective recognition is the baseline capability; agent-specific behavior is the substance of it.
Mobient’s NASA SBIR Phase I proposal applies these primitives to intravehicular logistics across four operational modes: autonomous cargo transport, crew-carried locomotion and task transport, MPM-assisted oversized payload handling, and external-tow hold-and-guide.
Four coupling primitives
At the highest level, Agentic Gravity Synthesis is a family of four coupling primitives:
Reactive Surface Locomotion
Timed capture-hold-release coupling for crew gait.
Magnetic Valley Shaping
Propulsive traveling-valley coupling for passive SLED transport.
Static Coupling
Non-propulsive hold coupling for station-keeping and staging.
Lane-Keep Coupling
Non-propulsive alignment and orientation control under external translation.
Together, these primitives let the same active-surface infrastructure support crew walking, cargo transport, task station-keeping, oversized payload handling, and externally towed heavy-infrastructure movement.
1. Reactive Surface Locomotion
Reactive Surface Locomotion (RSL) is timed capture-hold-release coupling for human gait.
In microgravity, handrails work but occupy the hands and force locomotion through the upper body. Passive magnetic boots impose a peeling penalty. Active boot concepts put mass, batteries, and controls on the moving foot.
RSL inverts that. Active hardware lives in the surface; the crew member wears passive footwear with compatible magnetic interfaces. As the foot approaches, the tile senses approach geometry, identifies or classifies the agent, pre-biases the relevant coil neighborhood, captures on contact, holds through stance, and tapers release through toe-off. If identity is delayed or unresolved, the surface falls back to a conservative default rather than requiring identity as a sole gate. The crew member experiences a predictable stance affordance without the peeling sensation of traditional magnetic boots.
RSL is the human-facing primitive. It gives the crew something gravity-limited environments normally remove: a floor that supports walking, stopping, bracing, and working hands-free. It is also the primary mechanism for crew-carried logistics. A crew member carrying tools, samples, or small payloads can translate without reaching for handrails.
2. Magnetic Valley Shaping
Magnetic Valley Shaping (MVS) is propulsive traveling-valley coupling for passive SLED cargo transport.
Cargo rides on a Surface-Linked Equipment Deck (SLED) with high-resistivity soft-magnetic coupling features. The active surface modulates coil currents across adjacent tiles to create a traveling magnetic potential valley under the SLED. The SLED is pulled along by the moving valley.
The architectural inversion is that the cargo interface is fully passive: it carries no onboard propulsion, no navigation, no coupling intelligence, and no electronics of any kind. Route planning, coil timing, tile-to-tile handoff, multi-agent reservations, collision avoidance, thermal governance, and stability gating all live in the infrastructure. Every cargo robot would otherwise need its own propulsion, batteries, navigation, and service burden; MVS supplies mobility as an infrastructure service to passive payloads.
MVS is the primitive for autonomous logistics and passive SLED transport. Any payload atop a SLED-class interface (cargo, samples, spares, waste, equipment, or robot chassis) can be moved across the surface by the traveling-valley mechanism. It is distinguished from Lane-Keep Coupling by the source of translation. In MVS, the surface supplies the motion.
3. Static Coupling
Static Coupling is non-propulsive hold coupling that keeps a recognized agent or payload fixed relative to the surface.
This is the foundational primitive. It is the programmable substitute for the holding function that friction and gravity provide on Earth.
- For a crew member: provides station-keeping: stop walking and the surface holds the stance.
- For a SLED: provides hold-down at staging points.
- For tools or equipment interfaces: provides local restraint without installing dedicated mechanical hardware everywhere a task might occur.
- For larger payloads: provides drift prevention during adjacent operations.
Static Coupling is not a gait cycle (RSL), does not move the payload (MVS), and is not path-following during motion (Lane-Keep). It is a hold state.
This primitive is what lets the active surface replace much of the fixed-position restraint hardware that currently occupies gravity-limited interiors. The surface can create a hold point wherever a compatible agent stops and the safety logic authorizes coupling.
4. Lane-Keep Coupling
Lane-Keep Coupling is non-propulsive coupling that keeps a recognized payload bounded, aligned, and oriented within a commanded lane while an external source supplies translation.
This primitive is specifically not MVS. In MVS, the surface supplies translation. In Lane-Keep Coupling, translation comes from elsewhere: winch, capstan, crawler, Mobile Push Mount (MPM) force, crew-applied force, or another external traction source. The infrastructure may still coordinate and execute that translation autonomously; what defines Lane-Keep is that the translation is non-MVS, not that it forgoes autonomy. The surface supplies lateral alignment, anti-drift authority, yaw and orientation control, bend guard, and hold-down during externally driven motion.
The analogy is lane-keep assist in a vehicle. The drivetrain supplies forward motion; lane-keeping supplies correction that keeps the vehicle bounded and aligned. In this architecture, the external tow or MPM supplies translation while the active surface supplies the lane-keeping coupling that prevents drift, yaw, or boundary violation.
Lane-Keep Coupling is especially important for heavy payloads that exceed efficient electromagnetic propulsion envelopes. At sufficient mass or geometry, using the surface as the main propulsor becomes inefficient. The surface can still provide substantial value by keeping the payload coupled, aligned, and safe while another source moves it.
Curves and corners are handled compositionally. A cornering operation combines external tow for forward translation, MPMs for lateral and yaw authority, Lane-Keep Coupling for lane adherence, and Static Coupling for pause or re-index states. Lane-Keep is not required to do curvature alone.
Mapping primitives to operational modes
The four operational modes are compositions of primitives, not separate behaviors.
| Mode | Primary primitives | Operational role |
|---|---|---|
| Mode 1 – Autonomous Cargo Transport | MVS + Static + Lane-Keep | Passive SLED cargo moves between stowage, staging, and work areas while the surface supplies route planning, tile-to-tile handoff, stability governance, and hold states. |
| Mode 2 – Crew-Carried Locomotion and Task Transport | RSL + Static | Crew members walk, stop, brace, and work hands-free while the surface provides capture-hold-release gait support and station-keeping. |
| Mode 3 – MPM-Assisted Oversized Payload Handling | MVS + Lane-Keep + Static | MPMs expand the Dynamic Stability Envelope by adding footprint, lateral authority, and center-of-mass management for oversized or unstable payloads. |
| Mode 4 – External-Tow Hold-and-Guide | Lane-Keep + Static | External traction supplies translation while the surface provides hold-down, alignment, bend guard, lane adherence, and stop states. |
Digital Twin Advancement Plan
The current replay viewer demonstrates deterministic run → log → replay behavior for selected DSE and thermal-governance scenarios. Phase I advances this substrate from replayable experiment traces into a breadboard-correlated reference-layout twin, and later into an operator-facing map and permissioned admin interface for live surface state, safety governance, maintenance, and audit.
| Advancement | Purpose | Role in the twin |
|---|---|---|
| Layer 1 Operational map and surface state | ||
| Reference corridor layout | Extend the twin from isolated scenarios to an L-shaped corridor-scale virtual layout with tile adjacency, route reservation, handoff logic, and multi-agent coordination. | Shows how AGS primitives compose across a spacecraft interior path before full multi-tile hardware is built. |
| Live map / operational state view | Add a 2D operator-facing surface map showing agents, tile state, coil activation, route reservations, DSE margins, current LED guidance patterns, commanded hold/release zones, and replay position over time. | Lets operators see where the surface is guiding, warning, holding, throttling, releasing, refusing, or reserving authority. |
| Fault, outage, and degraded-state overlays | Display unavailable or derated floor regions, including offline tiles, disabled coils, failed or removed LRUs, communication faults, unrecognized-agent zones, thermal derates, maintenance lockouts, coolant-unknown regions, and safety keep-outs. | Allows routes, LED guidance, reservations, and DSE decisions to account for unavailable or degraded floor authority. |
| Surface LED and crew-facing state visualization | Represent commanded LED patterns, caution/warning states, route cues, hold zones, release zones, maintenance lockouts, keep-outs, and emergency-state indications. | Connects the internal twin state to what crew, robots, and operators would actually see on the active surface. |
| Agent, route, and reservation tracking | Track recognized crew footwear, SLEDs, robots, MPMs, equipment bases, active reservations, route segments, handoff states, and DSE authorization/refusal decisions. | Turns the twin from a replay chart into an operational traffic and surface-state model. |
| Layer 2 Asset health and configuration | ||
| Wear, service, and tile-history state | Maintain tile and LRU history including cumulative coil activity, duty history, thermal cycling, wear indicators, service events, replacement dates, inspection notes, and fault recurrence. | Supports maintenance planning, degraded-mode decisions, reliability tracking, and post-event diagnosis. |
| Configuration, calibration, and identity state | Track LRU identity, installed position, orientation, connector identity, firmware version, calibration constants, coil maps, driver limits, sensor profiles, and commissioning status. | Ensures that the twin reflects the actual installed tile, not just an abstract nominal tile. |
| Layer 3 Sensorless thermal governance | ||
| Coil-level thermal governance | Extend the thermal model into a sensorless per-coil thermal map, estimating local headroom, cooldown state, heat accumulation, PCM/coolant coupling, and derating needs without requiring dedicated per-coil temperature sensors. | Supports safe coil scheduling by adjusting current, duty cycle, pulse timing, coil selection, cooldown intervals, route allocation, or degraded-mode behavior when modeled thermal headroom is limited. |
| Local and installed-position thermal context | Represent thermal state across coils, coil groups, LRUs, tile quadrants, coolant-coupled regions, PCM-adjacent regions, and installed LRU positions. | Captures the fact that nominally identical coils or LRUs may behave differently depending on geometry, adjacent-coil activity, coolant proximity, PCM adjacency, and installed tile position. |
| Layer 4 Audit, admin, and validation readiness | ||
| Audit, replay, and event history | Preserve deterministic replay logs, DSE decisions, thermal derates, fault events, operator actions, route reservations, LED-state changes, and hardware-state transitions. | Provides a reproducible explanation of what the surface did, why it did it, and what state it believed the system was in at the time. |
| Admin / maintenance console | Feed the operational twin into a permissioned interface for commissioning, tile/LRU registration, calibration review, service history, fault triage, lockout control, derate review, replay inspection, and maintenance planning. | Gives operators and maintainers a controlled interface for commissioning, maintaining, reviewing, and constraining the surface without treating the public replay viewer as the administrative tool itself. |
| Phase II multi-tile validation readiness | Package replay logs, hash reports, calibration constants, tolerance bands, data schemas, and test matrices for future hardware expansion. | Creates the handoff from Phase I breadboard correlation to expanded Phase II multi-tile, hardware-correlated validation. |
In this roadmap, the visible replay viewer is the public evidence layer. The mature digital twin becomes both an operator map and an administrative substrate: it shows what the surface is doing, what the surface believes about its own health, and what maintenance or safety constraints govern future commands.
Physics basis of the current DSE twin
The current DSE engine treats payloads as scenario-defined control objects, not decorative CAD bodies. Each replay configuration supplies the properties the floor must govern: effective mass, center-of-mass height, support geometry, CG uncertainty, commanded demand, and local floor authority.
Those properties feed a support-envelope calculation: mass and center-of-mass height increase the overturning demand a maneuver creates, while available support geometry increases the stabilizing lever arm. The result is a deterministic estimate of how much floor authority a command requires, whether it is admitted, and how much demand must be throttled. Tow traces are governed by external traction and attachment geometry; the human/RSL trace is an impedance/contact-feel model rather than a body-mass simulation; thermal traces govern a heat-token budget.
The charts then visualize the logged result: commanded demand, admitted fraction (alpha), DSE band, authorization, cost, normal force, force-rate, and for thermal– heat-budget accounting.
Phase I will mature the replay framework toward a profile-driven twin in which tile hardware facts, environmental boundary conditions, and calibration packs are versioned inputs to the deterministic engine rather than hidden constants.
Physics note: how geometry becomes force
A simplified reading: required hold-down force rises with mass, commanded acceleration, and center-of-mass height, and falls as the support lever arm grows. The current DSE traces govern inertial overturning and electromagnetic hold/guide authority rather than modeling terrestrial weight or friction.
F_req = SF · m⁺ · a_e · h⁺ / (η · d_edge_min)
Thesis
The surface decides, moment by moment, how gravity’s normal surface work relationship is reconstructed for each recognized agent: walk, hold, move, lane-keep, stabilize, release, or refuse. Agentic Gravity Synthesis moves gravity’s work-enabling surface relationship out of the vehicle-wide environment and into programmable, per-agent surface interactions. It chooses what to synthesize, for whom, and when.