---
title: 'DMPC-SWARM: Distributed MPC for UAV Swarms'
url: https://www.emergentmind.com/topics/dmpc-swarm
type: topic
---

# DMPC-SWARM: Distributed MPC for UAV Swarms

DMPC-SWARM denotes a class of distributed model predictive control architectures for UAV and multirobot swarms in which coupled motion-planning, collision-avoidance, and formation-control problems are solved in receding-horizon form through local or partially local optimizations that exchange predicted trajectories, state estimates, or consensus variables. In the recent arXiv literature, closely related methods described under the DMPC-SWARM label include event-triggered ground-offloaded planning for UAV swarms [2206.11020], online trajectory generation with on-demand collision avoidance [1909.05150], decentralized agile quadrotor collision avoidance based on ORCA and flatness-based MPC [1909.03961], communication-efficient trajectory-prediction-assisted DMPC [2401.12852], distributed learning-based predictive control with explicit closed-loop policies [2412.19669], and a nano-UAV implementation combining event-triggered off-board computing with a stateless wireless mesh and message-loss recovery [2508.20553]. This suggests that DMPC-SWARM is best understood as a research lineage rather than a single canonical algorithm.

## 1. Core formulation and system models

At its most standard, DMPC-SWARM assigns each agent \(i\) a local prediction model, a finite horizon, a stage cost, and state/input/safety constraints, then executes only the first control action before shifting the horizon. In the event-triggered UAV-swarm formulation, each vehicle is modeled by the linear time-invariant discrete-time system
\[
x_i(k+1)=A\,x_i(k)+B\,u_i(k),
\]
with
\[
x_i(k)=\begin{bmatrix} p_i(k)\ v_i(k)\ y_i(k) \end{bmatrix},
\]
where \(p_i\in\mathbb R^3\) is position, \(v_i\in\mathbb R^3\) velocity, and \(y_i\in\mathbb R^{n-6}\) additional internal states [2206.11020]. The local decision variables over horizon \(H\) are predicted state and input sequences \(X_i(k)\) and \(U_i(k)\), and the optimization minimizes a quadratic tracking-and-effort objective subject to dynamics, box constraints, a terminal constraint \(x_i(k+H\mid k)\in\mathcal X_{f,i}\), and pairwise collision-avoidance constraints [2206.11020].

A second widely used formulation parameterizes future trajectories directly. In the online multi-robot motion-planning variant, each agent uses a linear discrete-time trajectory-tracking model
\[
x_i[k+1]=A_i\,x_i[k]+B_i\,u_i[k],\qquad
x_i[k]=\begin{pmatrix}p_i[k]\\v_i[k]\end{pmatrix}\in\mathbb R^6,
\]
and samples the continuous control trajectory on the horizon by concatenating \(l\) Bézier curves of degree \(p\). The control points \(\mathcal U_i\) become the QP decision variables, while physical limits are imposed by sampling the curves and their derivatives up to order \(r\) and encoding the bounds as linear inequalities on \(\mathcal U_i\) [1909.05150].

For agile quadrotors, the modeling layer can be richer. The decentralized collision-avoidance method DCAD begins from the full 12-dimensional nonlinear quadrotor dynamics and then exploits differential flatness with flat outputs \(\zeta=[x,y,z,\psi]^T\). Through exact feed-forward linearization, the dynamics are transformed into the linear decoupled form
\[
\dot\xi=A\,\xi+B\,\nu+c,
\]
which is then discretized and embedded in an MPC with velocity, acceleration, jerk, and collision constraints [1909.03961]. This formulation is explicitly downwash conscious and incorporates Kalman filtering to address uncertainty in sensed position and velocity data [1909.03961].

Later work generalizes beyond repeated online QP solves. Distributed learning-based predictive control formulates the multirobot problem as a nonlinear coupled finite-horizon MPC with dynamics
\[
x_{i,k+1}=f_i(x_{N_i,k})+g_i(x_{i,k})u_{i,k},
\]
local stage costs \(\ell_i(x_{N_i},u_i)=\|x_{N_i}\|_{Q_i}^2+\|u_i\|_{R_i}^2\), and coupling sets \(E_i=\{x_{N_i}\mid g_{ij}(x_i,x_j)\le0\ \forall j\in N_i\}\), then learns explicit closed-loop policies to avoid numerical optimization during deployment [2412.19669]. A plausible implication is that DMPC-SWARM now spans both optimizer-centric and policy-centric implementations.

## 2. Collision avoidance, constraint handling, and local objectives

The principal technical distinction among DMPC-SWARM variants lies in how they encode inter-agent safety. In the event-triggered UAV architecture, pairwise avoidance is enforced through time-variant buffered Voronoi cells (TV-BVC). For each sample \(\ell\), the constraint
\[
n_{ij}(\ell)^\top\bigl(p_j(k+\ell\mid k)-p_i(k+\ell\mid k)\bigr)\ge\tfrac{1}{2}\bigl(d_{\min}+\|n_{ij}(\ell)\|\bigr)
\]
is built from previously predicted relative positions, where \(d_{\min}\) is the minimum allowed inter-UAV distance. By enforcing these linear constraints at sufficiently many points along the horizon, continuous-time collision-free operation is guaranteed [2206.11020].

The online trajectory-generation approach replaces full Voronoi partitioning by an on-demand mechanism. Each agent first detects whether a predicted collision with neighbor \(j\) exists within the horizon through the earliest index
\[
k_{c,ij}=\min\Bigl\{k:\,\bigl\|\Theta^{-1}\bigl(p_i[k|k_t-1]-p_j[k|k_t-1]\bigr)\bigr\|_2<r_{\min}\Bigr\},
\]
defines an activation set \(\Omega_i\) of imminent conflicts, and adds linearized separating-hyperplane constraints only for \(j\in\Omega_i\). Slack variables \(\varepsilon_{ij}\le0\) soften these constraints, while the local objective combines terminal position error, smoothness or energy regularization, and violation penalties [1909.05150]. The paper explicitly states that this is far less conservative than full buffered-Voronoi methods [1909.05150].

In DCAD, collision constraints are built from Optimal Reciprocal Collision Avoidance (ORCA) half-spaces. At each prediction step, agent \(A\) enforces
\[
(v^A-v^B)\cdot n \ge \frac{r_A+r_B-n\cdot(p^B-p^A)}{\tau},
\]
and extends the geometry through a sphere-oriented-ellipsoid Minkowski model so that a higher vehicle’s downwash region is respected. The method then inflates these shapes by the maximal standard deviation of the neighbor’s position covariance from a Kalman filter [1909.03961]. This is a notably different safety mechanism from TV-BVC or on-demand hyperplane constraints, but it remains compatible with linear or linearized MPC.

Communication-constrained DMPC introduces a broader constraint stack. In the cluttered-environment formulation, the local QP includes dynamics feasibility, Bézier continuity, agent-agent collision avoidance, cohesion, and agent-obstacle avoidance:
\[
A_{saf-agent,i}^k(\tilde U_{-i}^k)[w_i^{k\,T},\varepsilon_i^{k\,T}]^T\le b_{saf-agent,i}^k(\tilde U_{-i}^k),
\]
\[
A_{coh,i}^k(\tilde U_{-i}^k)[w_i^{k\,T},\delta_i^{k\,T}]^T\le b_{coh,i}^k(\tilde U_{-i}^k),
\]
\[
A_{saf-obs,i}[w_i^{k\,T},\zeta_i^{k\,T}]^T\le b_{saf-obs,i},
\]
with nonnegative slacks \(\varepsilon_i^k,\delta_i^k,\zeta_i^k\). Its stage cost combines migration to the goal, control effort, inter-agent and agent-obstacle repulsion, and cohesion [2401.12852].

A common misconception is that DMPC-SWARM always relies on a single safety formalism. The literature instead shows several distinct mechanisms—TV-BVC, on-demand linearized collision constraints, ORCA half-spaces with downwash geometry, soft-safe-cohesive Bézier constraints, and barrier-augmented safe learning—embedded in the same receding-horizon coordination paradigm [2206.11020].

## 3. Distributed coordination, communication, and triggering

Distributed coordination in DMPC-SWARM ranges from purely peer-to-peer trajectory exchange to architectures that offload planning to ground-based computation units. In the online trajectory-generation framework, each agent broadcasts its predicted trajectory \(\Pi_i[k_t]\) to its neighbors, collects \(\{\Pi_j[k_t]\}\), solves its local QP in parallel, and re-broadcasts its updated prediction. Because each QP only depends on one-hop neighbors in \(\Omega_i\), the computation parallelizes naturally [1909.05150]. DCAD adopts a similar decentralized pattern, exchanging state estimates, covariances, and optionally the first few steps of predicted trajectories with neighbors within a sensing radius \(R_{\rm sense}\), with communication rates matched to the MPC rate [1909.03961].

The event-triggered UAV-swarm architecture of 2022 introduces a different organization. There are \(M\) stationary ground-based computation units and \(N\) UAVs, with \(M<N\), and communication proceeds in synchronous rounds of length
\[
T=T_{\rm calc}+T_{\rm com}.
\]
During the communication slot, all agents broadcast their most recent state or planned trajectory in a many-to-all flood so that every CU and UAV begins the next round with a consistent global view of trajectories. An event trigger then selects which \(M\) of the \(N\) UAV trajectories should be replanned in the coming round [2206.11020]. The Priority-Based Trigger assigns each UAV a scalar score
\[
g_i(k)=\alpha_1\|p_i(k)-p_{i,\rm target}\|^2+\alpha_2T_i(k)-\alpha_3\sum_{j\neq i}w_{ij}(k),
\]
combining tracking error, time since last replanning, and a blocking measure [2206.11020].

The 2025 nano-UAV implementation retains the same event-triggered off-board idea but embeds it in a specific communication stack. All UAVs and compute units are synchronized into fixed-length rounds by the Mixer wireless protocol, with a computation phase and a communication phase. Each CU updates “information-trackers” \(\mathcal D_{i,w}(k)\), runs the event trigger if all trackers are up-to-date, solves one local MPC, and broadcasts a trajectory message. If any tracker is deprecated because of missing messages, the CU enters message-loss-recovery mode and requests the missing trajectory. Priority values can be Round-Robin, Distance-based, or Hybrid, and they are unified by max-consensus across CUs [2508.20553].

Self-triggering provides a further reduction in communication. For asynchronous nonlinear multi-agent systems with bounded delays, each agent solves a local min-max MPC at its own triggering instants, evaluates value functions \(V_{i,N}^H\) for candidate triggering intervals \(H\in\{1,\dots,\bar H\}\), and chooses the largest interval whose worst-case cost does not exceed the one-step update cost:
\[
H^{i*}(t_k^i)=\max\Bigl\{H:\,
V_{i,N}^{H}(x_i(t_k^i),\bm x_{-i}(t_k^i))
\le
V_{i,N}^{1}(x_i(t_k^i),\bm x_{-i}(t_k^i))
\Bigr\}.
\]
The agent then broadcasts a predicted trajectory and waits until the next triggering instant \(t_{k+1}^i=t_k^i+H^{i*}(t_k^i)\) [2106.05358].

These designs show that “distributed” does not imply a single networking pattern. Depending on the formulation, DMPC-SWARM may mean local peer-to-peer exchange, centralized many-to-all flooding with off-board compute, asynchronous self-triggering, or a hybrid architecture with consensus over trigger priorities and message-loss recovery [2508.20553].

## 4. Theoretical guarantees

The strongest theoretical results in DMPC-SWARM concern recursive feasibility and collision avoidance. In the event-triggered ground-offloaded formulation, Lemma 1 proves safety at sample points, Theorem 1 establishes recursive feasibility under the terminal constraint and TV-BVC conditions even if some agents are not replanned, and Theorem 2 shows continuous-time collision avoidance when the TV-BVC sampling grid is dense enough and
\[
d_{\min}-\Delta p_{\max}(T)\ge0.
\]
The proof strategy combines standard MPC Lyapunov-like arguments with buffered Voronoi geometry [2206.11020].

The nano-UAV implementation extends these guarantees to more realistic communication conditions. Under one-round communication delay, arbitrary message losses detected through information-trackers, and bounded tracking error, Lemma 1 proves correctness of information-trackers, Lemma 2 proves tracker agreement across computation units, Theorem 1 proves recursive feasibility and collision-free reference generation, and Theorem 2 establishes actual continuous-time separation:
\[
\forall t\ge0,\ \forall i\neq j:\quad
\big\|\Theta^{-1}(p_j(t)-p_i(t))\big\|_2\ge d_{\min},
\]
provided the tracking error bound satisfies
\[
\Delta d_{cont}\le \hat d_{\min}-(d_{\min}+\Delta d_{\min}).
\]
The paper explicitly states that safety is maintained despite one-round delay, arbitrary message losses, and bounded tracking error [2508.20553].

In the asynchronous self-triggered min-max setting, a consistency constraint restricts the deviation between a newly optimized trajectory and the last broadcast one,
\[
\bigl\|x_i(s\mid t_k^i)-x_i^b(s+t_k^i-t_{k-1}^i\mid t_{k-1}^i)\bigr\|\le\Delta_i,
\]
so that previously broadcast predictions remain usable even under time-varying delays. Under Lipschitz dynamics, bounded disturbances and delays, and a robust positive-invariant terminal set, Theorem 1 guarantees recursive feasibility and input-to-state practical stability at update instants [2106.05358].

Optimization-theoretic guarantees also appear in distributed solution methods. For cooperative formation control, ADMM is globally convergent for convex QP formulations, while decentralized SQP for non-convex QCQP subproblems enjoys local convergence under standard SQP assumptions, including LICQ, strict complementarity, and a strong second-order sufficient condition [2301.07960]. In the spatial-temporal D-PDDP framework, ADMM primal and dual residuals converge to zero under mild convexity assumptions on the constraint-indicator terms plus local fast convergence of PDDP, and spectral tuning of the penalty parameter is proposed to reduce the number of algorithmic iterations [2510.17541].

Learning-based variants preserve an MPC-style stability narrative. Distributed policy learning selects terminal cost weights \(P_i\) through a Lyapunov equation and imposes a baseline-upper-bound condition
\[
J(e(k))\le J^b(e^b(k)),\ \forall k,
\]
so that policy-learning updates decrease the cost until convergence and closed-loop asymptotic stability follows [2412.19669]. A plausible implication is that DMPC-SWARM now includes both classical recursive-feasibility proofs and stability arguments mediated by learned policy approximators.

## 5. Empirical performance and hardware realizations

Reported results show that DMPC-SWARM has progressed from simulation-only studies to hardware-in-the-loop and fully distributed swarm experiments. The numerical evidence is heterogeneous because the papers evaluate different dynamics, communication models, and optimization back-ends, but several recurring themes are clear: communication reduction, reduced conservatism relative to buffered-Voronoi baselines, and real-time feasibility on swarms ranging from a few robots to tens of thousands of simulated agents.

| Variant | Setup | Reported outcome |
|---|---|---|
| Event-triggered off-board ET-DMPC [2206.11020] | Up to \(N=25\), \(M=10\), \(5\times5\times5\) m\(^3\) arena | \(60\%\) reduction in communication load, \(\approx60\%\) savings in cumulative computation time, \(>98\%\) target reach vs. \(92\%\) round-robin |
| On-demand DMPC [1909.05150] | Up to 60 agents in \(18\) m\(^3\) | \(>90\%\) success rates up to 30 agents; average transition times reduce by \(\sim50\%\) vs. BVC |
| DCAD [1909.03961] | 8 agents, up to \(7\) m/s nominal speed | 0 collisions up to \(4\) m/s; at \(7\) m/s, 21/250 collisions vs. ORCA 151/250 and AVO 128/250 |
| Prediction-assisted DMPC [2401.12852] | Funnel environment, \(n\in\{6,\dots,14\}\) | VAE+EG+KKT total cost \(\approx210.8\) k vs. oracle DMPC \(\approx208.4\) k; communication load cut by up to \(50\%\) |
| DMPC-Swarm on nano UAVs [2508.20553] | 16 Crazyflie 2.1, 3 laptops, real wireless mesh | \(>99.999\%\) many-to-all delivery per round; formation change \(\approx15\) s with 1 CU and \(\approx7\) s with 3 CUs |

Hardware validation is a consistent feature. The online trajectory-generation method was implemented in C++/ROS on an 8-core Xeon workstation, replanning at \(5\) Hz and sending commands at \(20\) Hz, with qpOASES solve times from \(\sim3\) ms for 2 agents to \(\sim28\) ms for 20 agents. Experiments with 2–20 Crazyflie 2.0 quadrotors validated safe online transitions, and a “hula-hoop” scenario with 10 drones passing through a 30 cm aperture was completed in 28 s [1909.05150].

The 2022 event-triggered off-board framework was further tested in hardware-in-the-loop with six Crazyflie quadcopters tracked by a motion-capture system and controlled via a single on-ground computation unit using a linear triple-integrator model. The reported trajectories were collision-free, smooth, and easily executed by the on-board PD controllers [2206.11020].

DCAD reports mean per-agent compute time of approximately \(5\) ms on a modern CPU core with \(N=10\) and up to 10 neighbors, versus approximately \(16\) ms in the compared nonlinear MPC implementation, while maintaining trajectory length and time-to-goal within \(<1\%\) of the obstacle-free optimal [1909.03961]. Cooperative DMPC via decentralized real-time optimization gives median solve times of 6.6 ms per MPC step for a convex chain-formation QP and 33.8 ms for a non-convex minimum-distance problem, with communication accounting for \(60\)–\(85\%\) of ADMM or dSQP execution time [2301.07960].

The largest simulated scale appears in distributed learning-based predictive control, where online learning time per step grows approximately \(O(M)\) from 0.02 s for \(M=2\) to 14.6 s for \(M=10\,000\) on a laptop, while direct policy deployment ranges from \(2\times10^{-5}\) s to 0.28 s. The same work reports deployment to robot populations up to \(10\,000\) units, Gazebo multirotor experiments with \(M=6,18,40\), and \(5\)–\(20\times\) speed-up over classical DMPC implementations using IPOPT or fmincon [2412.19669].

## 6. Variants, limitations, and current directions

DMPC-SWARM now encompasses several distinct research directions. One direction seeks lower communication and computation through event-triggering, off-board replanning, and stateless many-to-all wireless flooding [2206.11020]. A second direction reduces conservatism through on-demand collision constraints and sparse neighbor interactions rather than full buffered partitions [1909.05150]. A third direction incorporates richer dynamics and uncertainty models, as in flatness-based quadrotor MPC with downwash-aware ORCA constraints and Kalman-filter-based covariance inflation [1909.03961]. More recent directions couple DMPC with learned trajectory prediction under limited communication [2401.12852] or replace online numerical optimization by distributed actor-critic policy learning with barrier-augmented safety shaping [2412.19669].

The literature also identifies several limitations. The online trajectory-generation study notes that, as agent density exceeds approximately 30 in \(18\) m\(^3\), collisions become more frequent and relaxations larger, motivating hierarchical or waypoint planners or stronger constraint tightening; it also explicitly assumes perfect, delay-free neighbor broadcasts [1909.05150]. The nano-UAV implementation states that safety cannot be guaranteed if the tracking error bound is violated by extreme gusts, because the analysis assumes \(\|p_i(t)-\hat p_i(t)\|_2\le\Delta d_{\min}\) [2508.20553]. The decentralized formation-control implementation emphasizes that communication latency becomes critical for larger swarms and denser graphs, and observes experimentally that communication may dominate runtime [2301.07960].

A further point of variation concerns where computation resides. Some formulations run one local QP per agent in parallel on-board or on workstation-class processors [1909.05150], whereas others deliberately move heavy computation off the UAV to reduce on-board weight and power consumption and to introduce active redundancy against computation-unit failures [2206.11020]. By contrast, distributed learning-based predictive control argues that scalability for nonlinear multirobot systems may require explicit policies rather than periodic numerical optimization at all [2412.19669]. This suggests that the main contemporary debate is no longer whether DMPC can coordinate swarms, but which combination of optimization, communication, prediction, and learning yields the best trade-off between safety guarantees, scalability, and deployability.

The most recent extensions push DMPC-SWARM toward broader optimal-control formulations. Distributed spatial-temporal trajectory optimization introduces free final time, safe-copy variables, ADMM-based consensus, and Parameterized Differential Dynamic Programming as a local solver, together with adaptive spectral-gradient tuning of the ADMM penalty parameter [2510.17541]. Taken together with prediction-assisted and policy-learning variants, these developments indicate that DMPC-SWARM is evolving from a narrowly defined local-QP coordination method into a broader framework for safety-critical, communication-aware, and increasingly heterogeneous swarm control.

Source: https://www.emergentmind.com/topics/dmpc-swarm