EMFF: Electromagnetic Formation Flight

• Electromagnetic Formation Flight (EMFF) is a satellite coordination method that uses programmable magnetic dipole interactions for formation-keeping with zero propellant.
• EMFF leverages frequency multiplexing of alternating magnetic fields to decouple intersatellite forces and enable decentralized, precise control.
• Experimental and simulation studies demonstrate centimeter-level accuracy and robust constraint enforcement in multi-satellite EMFF systems.

Electromagnetic Formation Flight (EMFF) refers to the control and coordination of multiple spacecraft using electromagnetic forces—specifically magnetic dipole–dipole interactions—rather than conventional fuel-based thrusters. By equipping each satellite with controllable electromagnetic coils, satellites in a swarm generate programmable magnetic fields that interact to produce inter-satellite forces and torques. These internal forces enable formation-keeping, reconfiguration, and even combined position–attitude control with zero propellant consumption. The intrinsic challenges of strong coupling, system nonholonomy, and severe actuation and safety constraints have led to the development of sophisticated modeling, control, and experimental technologies detailed below.

1. Physical Model and Nonholonomic Dynamics

In EMFF, each satellite is modeled as a rigid body equipped with tri-axial electromagnetic coils that generate a body-fixed magnetic dipole moment. The overall system is described by inertial positions $r_j \in \mathbb{R}^3$, body-attitude coordinates (e.g., Modified Rodrigues Parameters, $\sigma_j \in \mathbb{R}^3$), and angular velocities. The instantaneous force on satellite $i$ due to satellite $j$ is governed by the magnetic dipole–dipole formula:

$$F_{ij}(t) = \frac{3 \mu_0}{4\pi \|r_{ij}\|^4} \left[ (u_j \cdot \hat e_{ij})u_i + (u_i \cdot \hat e_{ij})u_j + (u_i \cdot u_j - 5(u_i \cdot \hat e_{ij})(u_j \cdot \hat e_{ij}))\hat e_{ij} \right]$$

with $r_{ij} = r_i - r_j$, $\hat e_{ij} = r_{ij}/\|r_{ij}\|$, and $u_i$ the net dipole moment of satellite $i$ (Kamat et al., 2024).

Crucially, as EMFF generates only internal forces, the total system angular momentum is conserved:

$$H_{total} = \sum_{j=1}^n \left[ m_j (r_j - r_1)\times \dot r_j + I_j \omega_j \right]$$

Enforcing $\sigma_j \in \mathbb{R}^3$0 yields a nonholonomic velocity-level constraint $\sigma_j \in \mathbb{R}^3$1 on the generalized velocity $\sigma_j \in \mathbb{R}^3$2, with $\sigma_j \in \mathbb{R}^3$3 built from position- and attitude-dependent terms. This constraint cannot be integrated to a position-only relation; the configuration space is thus nonholonomic and restricts achievable maneuvers to those permitted by available electromagnetic couplings (Takahashi et al., 6 Jan 2026).

System dynamics are cast in reduced order on the angular momentum–manifold by projecting onto the nullspace of $\sigma_j \in \mathbb{R}^3$4. The equations take the constrained Euler–Lagrange M–C form, with control input consisting of net electromagnetic forces and torques per satellite.

2. Alternating Magnetic Field Forces: Decoupling via Frequency Multiplexing

The control of EMFF is fundamentally complicated by the fully coupled nature of dipole–dipole interactions—each satellite's coil currents influence every other. To manage this, EMFF systems increasingly employ Alternating Magnetic Field Forces (AMFF) or frequency-multiplexed actuation. Each coil's total moment is synthesized as a sum of sinusoids, each at a distinct frequency $\sigma_j \in \mathbb{R}^3$5 unique to a satellite pair:

$\sigma_j \in \mathbb{R}^3$6

Over any period $\sigma_j \in \mathbb{R}^3$7 that is a common multiple of all $\sigma_j \in \mathbb{R}^3$8, cross-frequency terms time-average to zero. Thus, the averaged force between pairs $\sigma_j \in \mathbb{R}^3$9 depends only on their mutual amplitudes $i$0:

$i$1

This effectively decouples $i$2 pairwise interactions, enabling independent control of each "virtual spring" in the formation (Kamat et al., 8 Jan 2026, Kamat et al., 25 Aug 2025, Kamat et al., 2024).

Closed-form amplitude allocation laws are derived so that for each pair, the desired average force is realized through modulated sine amplitudes. In 1D testbeds, these amplitudes directly relate to command coil currents via $i$3, where $i$4 is coil turns and $i$5 area.

3. Control Law Design: Decentralized, Constrained, and Time-Varying Approaches

Several architectures have been proposed to implement formation control via EMFF:

• Time-Varying Kinematics Control: A nonlinear, time-varying homogeneous feedback law exploits high-frequency oscillatory terms to "average in" the necessary Lie-bracket directions omitted by nonholonomic constraints. This achieves both translational and full-attitude control—even in the absence of conventional attitude actuators (e.g., reaction wheels). The control law has the form $i$6, where $i$7 contains high-frequency terms tuned by a small $i$8 (Takahashi et al., 6 Jan 2026).
• Decentralized Linear-Consensus Laws: Each satellite measures the states of its neighbors and computes the required pairwise average force via double-integrator consensus or consensus-spring feedback. The force commands are realized by tuning dipole sine amplitudes per frequency (Kamat et al., 8 Jan 2026).
• Constrained Optimal and Barrier Function Methods: Practical EMFF systems encounter actuator saturation, collision avoidance, and inter-satellite velocity constraints. Constrained feedback architectures employ optimal control to generate unconstrained average forces, then project these via control barrier functions (CBFs)—or relaxed CBF compositions with soft-min (log-sum-exp) approximations—to ensure all state and input constraints are satisfied at each stage (Kamat et al., 25 Aug 2025, Kamat et al., 2024). The QP arising from this projection admits analytic solution due to the low problem dimension.

A typical control sequence is as follows: measure local state, compute the unconstrained desired force (e.g., via LQR or consensus), project to satisfy all constraints (barriers), allocate the necessary sine amplitudes, and update dipole moments for the next cycle.

4. Experimental and Numerical Demonstrations

EMFF technologies have advanced from purely theoretical treatment to hardware validation:

• The first experimental demonstration of a decentralized AMFF controller for three satellites was conducted on a 1D air-track platform. Each satellite was equipped with three orthogonal coils and local laser-range sensors, and inter-satellite spacing was regulated to prescribed setpoints under both attraction and repulsion maneuvers (Kamat et al., 8 Jan 2026).
• Experimental results for 2- and 3-satellite formations achieved <2 cm overshoot, settling times of 20–30 s, and strong correspondence (<15% error) with numerical simulations—for both open-loop and closed-loop behaviors.
• Numerical experiments with constrained CBF architectures confirm that minimum separation, velocity bounds, and current/power limitations are strictly enforced during station-keeping and dynamic reconfiguration for three or more agents (Kamat et al., 25 Aug 2025, Kamat et al., 2024).
• Time-varying kinematics controllers demonstrate that full position and attitude convergence can be achieved with no additional actuators, with simulations showing convergence from meter-scale to centimeter-scale errors in 200 s, and attitude errors from 0.5 rad to <0.01 rad in 150 s (Takahashi et al., 6 Jan 2026).

Paper	Platform	Key Result
(Takahashi et al., 6 Jan 2026)	Simulation (3D)	Full position+attitude, $i$91 cm error, no RWs
(Kamat et al., 8 Jan 2026)	1D air-track (exp)	3-sat AMFF, <2 cm overshoot, 20–30 s settling
(Kamat et al., 25 Aug 2025)/(Kamat et al., 2024)	Simulation/Numerics	CBF-constrained, all constraints enforced

5. Controllability, Stability, and Scalability

EMFF systems are nonholonomic due to global angular momentum conservation; this renders formation and attitude control non-integrable in general. Control-theoretic analysis demonstrates:

• Small-Time Local Controllability (STLC): Formal accessibility is established through Lie-bracket calculations and the Sussmann theorem, showing that the EMFF-reduced system is STLC on the angular momentum manifold. The "no bad bracket" condition is checked via Hall-basis computations (Takahashi et al., 6 Jan 2026).
• Local Exponential Stability: By constructing a Lyapunov candidate involving both translational and rotational (attitude) state energies, it is shown that, under sufficiently high-frequency and small-amplitude time-varying feedback, all errors decay exponentially in a local region (Takahashi et al., 6 Jan 2026).
• Scalability: The AMFF approach can in principle be extended to $j$0 satellites. This requires $j$1 distinct interaction frequencies, and the sampling period for amplitudes must be a common multiple of all involved frequencies. In practice, the number of satellites can be limited by update delays, available frequency bandwidth, and inter-satellite communication/estimation rates (Kamat et al., 8 Jan 2026).

6. Limitations, Practical Considerations, and Future Directions

Several limitations and open directions are recognized in the literature:

• Locality of Results: Most controllability and stability claims are local; the nonholonomic nature of the system and practical limits on oscillation amplitude ($j$2) confine the region of guaranteed convergence (Takahashi et al., 6 Jan 2026).
• Frequency Constraints: Sufficiently high AC frequencies are necessary to ensure that the time-averaged force approximation holds and cross-terms remain negligible; physical coil bandwidth and eddy current effects may limit this in practice (Kamat et al., 8 Jan 2026).
• Actuator Nonidealities: The coil-to-dipole mapping, unmodeled drag/perturbations, magnetic hysteresis, and hardware nonlinearities can degrade performance, as observed in increased experimental overshoot and RMS force relative to simulation (Kamat et al., 8 Jan 2026).
• Constraint Enforcement: The CBF approach successfully enforces collision, speed, and power constraints at the cost of moderation of the unconstrained optimal force command, but relies on sufficiently fast onboard computation for real-time QP solutions (Kamat et al., 25 Aug 2025, Kamat et al., 2024).

Future research directions include robustification against environmental disturbances, decentralized control and estimation architectures (each satellite leveraging local-only data), on-orbit demonstration in microgravity (e.g., MAGNARO, ETS-VII), and extensions to optimize for coil energy budgets via data-driven or model-predictive controllers (Takahashi et al., 6 Jan 2026).