MORPH: Self-Organising Multi-Robot Task Allocation via Neuroplasticity-Inspired Adaptive Topology

Xuezhi Niu; Didem Gürdür Broo

cs.ROSep 26, 2026

MORPH: Self-Organising Multi-Robot Task Allocation via Neuroplasticity-Inspired Adaptive Topology

Authors: Xuezhi Niu, Didem Gürdür Broo

Organizations: Department of Information Technology, Uppsala University, 751 05 Uppsala, Sweden

Abstract

Multi-robot task allocation (MRTA) in dynamic environments faces a fundamental tension: effective coordination requires learned structure, but that structure must adapt when conditions change. Existing methods resolve this by assuming prior task knowledge, a utility function, a cost matrix, or a trained policy making them brittle when deployed without such knowledge or when task distributions shift. We present MORPH(Multi-agent Online Rewiring through Plasticity-guided Hierarchy), a training-free MRTA framework where global allocation quality emerges from 4 local plasticity rules (synaptic, homeostatic, structural, and metaplasticity) applied to a directed pairwise preference matrix updated from runtime co-occurrence and task-completion feedback. MORPH requires no task model, no bid computation, and no offline training; response decisions use learned AGV-to-Picker preferences rather than a fixed proximity rule. Within the Gerkey-Mataric MRTA taxonomy, MORPH is the first method in the single-task, single-robot, instantaneous-assignment class to learn directed pairwise allocation preferences online. Evaluated on the TA-RWARE warehouse benchmark (8-24 agents, 4 maps, 800 steps per episode, 5 seeds), MORPH achieves 110% of all-to-all throughput at N=24 while using only 21% of possible coordination links as an efficiency advantage that grows monotonically with fleet size. Under spatial task distribution shift, MORPH degrades 3x less than proximity-based methods while its learned preferences remain uncorrelated with Manhattan distance. Systematic ablation confirms all four plasticity rules contribute measurably. Two allocation properties emerge without programming: cross-type preference dominance and progressive preference sparsification, mirroring the developmental refinement of biological neural circuits. Learned preferences are driven by task co-occurrence history, not spatial proximity.

Figures & tables

Rule	Plasticity form	Biological basis	Failure prevented
R1	Synaptic	Hebbian LTP + dopamine gating [ 13 , 5 ]	Stale preferences persist
R2	Homeostatic	Synaptic scaling [ 33 ]	Isolation / hub collapse
R3	Structural	Synaptogenesis + pruning [ 15 ]	Cold start; slow rewiring
R4	Metaplasticity	BCM-inspired dynamic thresholds [ 1 , 7 , 36 ]	Saturation; shift inertia

Table 1: MORPH plasticity rules: biological basis and allocation failure prevented.

Figure 1: TA-RWARE environment simulation at three fleet scales. Top row: initial state ( $t=0$ ) showing warehouse layout, shelf columns (orange squares), AGVs (hexagons), and Pickers (diamonds) before any preferences are established. Bottom row: mid-episode state ( $t=300$ ) showing MORPH’s self-organised coordination topology — green lines are stable W-weighted links between AGV–Picker pairs, red dashed lines are links being pruned by thresholds. Coordination complexity grows with fleet size. Full animations are available as supplementary material.

Parameter	Symbol	Value
Synaptic learning rate	$\alpha$	0.18
Base decay rate	$\gamma$	0.98
Homeostatic rate	$\beta$	0.04
Formation threshold (start/end)	$\theta_{\text{form}}^{0/T}$	0.75 / 0.45
Pruning threshold	$\theta_{\text{prune}}$	0.008
Target degree fraction	$\rho_{\text{target}}$	0.35

Table 2: MORPH parameter values, identical across all four scales.

Scale ( $N$ )	Method	Deliveries	%AP	Links (%max)
Tiny (8)	Proximity	$62.4\pm 4.0$	108%	22.2 (79%)
	TSG	$55.2\pm 5.8$	96%	2.6 (9%)
	MORPH	$\mathbf{62.4\pm 2.9}$	108%	27.0 (97%)
	Full Graph	$57.8\pm 18.0$	100%	28.0 (100%)
Small (12)	Proximity	$75.2\pm 1.6$	117%	36.2 (55%)
	TSG	$70.9\pm 2.1$	111%	3.7 (6%)

Table 3: TA-RWARE scale study (

T=800

, 5 seeds, mean

\pm

std). %AP = deliveries as % of Full Graph throughput. Links = mean active links (% of maximum possible). Proximity

r

calibrated per scale to match MORPH’s converged link count, giving Proximity the maximum possible link-budget advantage.

Figure 2: TA-RWARE scale study ( $T=800$ , 5 seeds, 4 scales). Top row: cumulative delivery curves (mean $\pm$ std shaded). Middle row: throughput retention as % of Full-Graph (left) and links used as % of maximum possible (right). Bottom row: grouped final-delivery bars at each scale. MORPH (blue) exceeds Full-Graph (grey) at every scale; link usage falls monotonically from 97% to 21% of maximum as $N$ grows from 8 to 24.

Figure 3: Plasticity rule ablation (4 scales, 5 seeds). Each ”v2 $-$ X” bar removes one mechanism from the full MORPH v2 system. MORPH v2 full (dark blue) leads all partial combinations at every scale. Removing neuromodulation (pink) produces the most consistent degradation; removing the reward gate (green) causes a 7-deliveries reduction at Small.

Figure 4: Adaptability under cyclical spatial task-distribution shifts at Medium scale ( $N=18$ , 20 seeds). Demand alternates across six 400-step phases (L–R–L–R–L–R), with dotted lines marking phase boundaries. Top: Delivery rate (rolling 50-step mean) across all six phases. Bottom: Recovery time on repeated regime returns (steps to reach 90% of baseline delivery rate), where lower values indicate faster recovery and a shrinking trend indicates memory compounding across cycles. From Phase 3 to Phase 6, MORPH v2 shows the smallest degradation ( $-7.4\%$ ), compared with Proximity ( $-25.4\%$ ), TSG ( $-14.6\%$ ), and MORPH v2 reset ( $-15.6\%$ ).

Condition	P1	P2	P3	P4	P5	P6	$\Delta_{6-3}$	r3	r4	r5	r6
Proximity	51.1	48.2	45.7	41.8	37.9	34.1	$-11.6$	45.0	57.9	96.0	83.5
TSG	49.9	46.0	44.6	41.4	38.8	38.1	$-6.5$	40.8	54.6	39.8	50.0
MORPH	50.0	49.1	47.5	46.1	45.2	44.0	$\mathbf{-3.5}$	36.9	23.1	35.7	17.7
MORPH (reset at shift)	50.0	48.9	48.2	46.5	43.9	40.8	$-7.5$	31.4	17.1	49.5	18.1

Table 4: Task-distribution shift results at Medium scale (

N=18

, 20 seeds). P1–P6 denote mean deliveries in six consecutive 400-step phases under alternating left/right package demand.

\Delta_{6-3}

reports the change from Phase 3 to Phase 6. r3–r6 report recovery timesteps for the corresponding return phases, where lower values indicate faster recovery.

Figure 5: $\mathbf{W}$ matrix evolution ( $N=18$ , seed=42, snapshots at $t=0,100,200,400,600,800$ ). Red dashed lines divide the AGV block (rows/columns 0–11) from the Picker block (12–17). By $t=100$ the AGV–Picker off-diagonal block is already structured; diagonal blocks remain near zero throughout, consistent with the task structure.

Explore similar work

Feb 20, 2026

FORMICA: Decision-Focused Learning for Communication-Free Multi-Robot Task Allocation

Most multi-robot task allocation methods rely on communication to resolve conflicts and reach consistent assignments. In environments with limited bandwidth, degraded infrastructure, or adversarial interference, existing approaches degrade sharply. We introduce a learning-based framework that achieves high-quality task allocation without any robot-to-robot communication. The key idea is that robots coordinate implicitly by predicting teammates' bids: if each robot can anticipate competition for a task, it can adjust its choices accordingly. Our method predicts bid distributions to correct systematic errors in analytical mean-field approximations. While analytical predictions assume idealized conditions (uniform distributions, known bid functions), our learned approach adapts to task clustering and spatial heterogeneity. Inspired by Smart Predict-then-Optimize (SPO), we train predictors end-to-end to minimize Task Allocation Regret rather than prediction error. To scale to large swarms, we develop a mean-field approximation where each robot predicts the distribution of competing bids rather than individual bids, reducing complexity from

O(NT)

O(T)

. We call our approach FORMICA: Field-Oriented Regret-Minimizing Implicit Coordination Algorithm. Experiments show FORMICA substantially outperforms a natural analytical baseline. In scenarios with 16 robots and 64 tasks, our approach improves system reward by 17% and approaches the optimal MILP solution. When deployed on larger scenarios (256 robots, 4096 tasks), the same model improves performance by 7%, demonstrating strong generalization. Training requires only 21 seconds on a laptop, enabling rapid adaptation to new environments.

Antonio Lopez, Jack Muirhead, Carlo Pinciroli

Dept. of Robotics Engineering, Worcester Polytechnic Institute, Worcester MA 01609, USA

Multi-Robot Task Allocation Multi-Agent Systems Decision-Focused Learning

Ask PDF

May 25, 2026

Acting on the Unseen: Communication-Free Collaborative Filtering for Decentralized Multi-Robot Task Allocation

Multi-robot task allocation usually assumes some combination of communication, known task models, or a coordinator. We study the opposite extreme, a regime common in practice but overlooked in theory, which we name Zero-Knowledge MRTA (ZK-MRTA): a robot team with no prior knowledge (no task models, not even the latent rank), no communication (no messages, no parameter sharing, no coordinator), and only a partial and privately-noisy view of a public stream of teammates' outcomes. A hidden low-rank structure governs which robot suits which task, and there are far more tasks than rounds, so most (robot, task) pairs are never attempted. Yet each robot can act well on tasks it never attempted, and onboard new tasks, by running online low-rank collaborative filtering over the broadcast (SwarmCF). The advantage over any structure-free learner is categorical, not a constant factor: a structure-free learner is provably at the prior-mean error floor on unseen pairs. We prove a matching per-robot sample complexity (Θ(d) versus Θ(n), in the rank d and the task count n), an anytime (cumulative-reward) separation under task scarcity, and a deterministic condition under which decentralized recovery from the masked broadcast is exact (validated empirically). Experiments quantify the value of the broadcast, a positive scaling law (per-robot unseen-pair skill rises with team size), and the strongest masking-robustness and anytime profile among low-rank methods, recovering most (about 80% on earned skill) of a centralized full-communication ceiling, and holding under capacity-1 contention and in a robotics-grounded sensing instance.

Alexander Apartsin, Yigal Meshulam, Yehudit Aperstein

Holon Institute of Technology, Holon, Israel. · Afeka Tel Aviv Academic College of Engineering, Tel Aviv, Israel

Multi-Robot Task Allocation Collaborative Filtering

Ask PDF

May 21, 2026

Auction-Consensus Algorithm with Learned Bidding Scheme for Multi-Robot Systems

Multi-Robot Task Allocation (MRTA) is a central challenge in decentralized multi-agent systems, where teams of robots must cooperatively assign and execute tasks under limited communication while optimizing global performance objectives. Auction-consensus algorithms, such as the Consensus-Based Bundle Algorithm (CBBA), provide scalable decentralized coordination with provable convergence, but rely on hand-crafted greedy scoring functions that often lead to suboptimal task allocations. This paper proposes a learning-enhanced auction-consensus framework in which CBBA's deterministic bidding mechanism is replaced by a neural bidding policy trained using reinforcement learning. Under a centralized training and decentralized execution paradigm, agents learn to compute task bids from partial local observations while retaining the standard auction and consensus phases for decentralized coordination. The learned bidding policy is trained using Proximal Policy Optimization with rewards shaped by proximity to globally optimal solutions obtained via mixed-integer linear programming. Multiple neural architectures are evaluated, including a Neural Additive Model, the Long Short-Term Memory (LSTM) model, and the Set Transformer Model. Experimental results across varying swarm sizes demonstrate that learned bidding policies can improve solution quality over classical CBBA while preserving decentralized execution. The proposed approach highlights the effectiveness of integrating reinforcement learning with classical distributed coordination algorithms, offering a scalable pathway toward higher-quality decentralized multi-robot task allocation.

Jose Rodriguez, Constantine Tarawneh, Sven Koenig +2

Department of Electrical and Computer Engineering, The University of Texas at Rio Grande Valley (UTRGV), Edinburg, USA · Department of Mechanical Engineering, UTRGV, Edinburg, TX, USA · Department of Computer Science, Donald Bren School of Information and Computer Sciences, University of California, Irvine +1

Multi-Robot Coordination Multi-Robot Task Allocation

Ask PDF