Event-Triggered Adaptive Consensus for Multi-Robot Task Allocation

Fidel Aznar; Mar Pujol; Álvaro Díez

cs.MAApr 8, 2026

Event-Triggered Adaptive Consensus for Multi-Robot Task Allocation

Authors: Fidel Aznar, Mar Pujol, Álvaro Díez

Organizations: Department of Computer Science and Artificial Intelligence, University of Alicante, Alicante, Spain

Abstract

Coordinating robotic swarms in dynamic and communication-constrained environments remains a fundamental challenge for collective intelligence. This paper presents a novel framework for event-triggered organization, designed to achieve highly efficient and adaptive task allocation in a heterogeneous robotic swarm. Our approach is based on an adaptive consensus mechanism where communication for task negotiation is initiated only in response to significant events, eliminating unnecessary interactions. Furthermore, the swarm self-regulates its coordination pace based on the level of environmental conflict, and individual agent resilience is managed through a robust execution model based on Behavior Trees. This integrated architecture results in a collective system that is not only effective but also remarkably efficient and adaptive. We validate our framework through extensive simulations, benchmarking its performance against a range of coordination strategies. These include a non-communicating reactive behavior, a simple information-sharing protocol, the baseline Consensus-Based Bundle Algorithm (CBBA), and a periodic CBBA variant integrated within a Behavior Tree architecture. Furthermore, our approach is compared with Clustering-CBBA (C-CBBA), a state-of-the-art algorithm recognized for communication-efficient task management in heterogeneous clusters. Experimental results demonstrate that the proposed method significantly reduces network overhead when compared to communication-heavy strategies. Moreover, it maintains top-tier mission effectiveness regarding the number of tasks completed, showcasing high efficiency and practicality. The framework also exhibits significant resilience to both action execution and permanent agent failures, highlighting the effectiveness of our event-triggered model for designing adaptive and resource-efficient robotic swarms for complex scenarios.

Figures & tables

Figure 1: This figure illustrates the robot’s high-level decision-making architecture, implemented as a Behavior Tree (BT). The tree processes logic from the root downwards and from left to right, enabling a prioritized system of behaviors. Composition Nodes are represented by circles, Leaf Nodes by rectangles, conditions by red, and actions by green or purple (grouped actions)

Figure 2: This figure presents a detailed view of the robot’s Behavior Tree (BT) architecture. It reveals the sophisticated contingency logic and resource control that enable local adaptation and failure recovery, making the overall system robust. Some previously simplified actions are expanded to show their complex sub-structures and the underlying intelligence of the robot’s behaviors. Composition Nodes are represented by circles, Leaf Nodes by rectangles, conditions by red, and actions by green or purple (grouped actions)

Parameter	Value	Description
Simulation Control
Simulation Duration ( $D$ )	3000s	Total simulation time for each trial.
Repetitions ( $T$ )	50	Number of trials conducted for each experimental configuration.
Environment
Arena Radius ( $R_{arena}$ )	32m	Radius of the circular operational area.
Number of Robots ( $N_{R}$ )	20	Total number of agents in the swarm for the baseline scenario.

Table 1: Baseline Experimental Parameters (‘base_R20_V100‘)

Figure 3: This figure shows the simulated Search and Rescue (SAR) environment at a specific time step (233 seconds). The circular grey dashed line represents the victims boundary. Robots are depicted as numbered circles with an outer ring indicating their assigned color (red, green, or blue). The colored areas extending from each robot represent their conical sensory perception field. Victims (tasks) are shown as numbered triangles, with their fill color indicating their wall color. Thin dashed lines trace the recent trajectories of the robots. A blue arrow originating from a robot indicates its current movement target.

Figure 4: This figure presents a simple behaviour tree architecture for our SAR problem without communication, called Tree . Composition Nodes are represented by circles, Leaf Nodes by rectangles, conditions by red, and actions by green or purple (grouped actions)

Figure 5: This figure presents a visual comparison of consensus mechanisms: The diagram highlights how CBBA-Tree (blue/dashed lines) and our CBBA-ETC (red solid) approach task acquisition in a behavior tree, showing periodic vs. event-driven logic

Algorithm	Victims Rescued	Messages Sent	CBBA Negotiations	Failed Rescues
c-cbba	21,856	104,535	10,687	16,547
tree	21,725	0	N/A	43,634
comm	21,930	435,457	N/A	39,193
cbba	21,756	271,415	61,685	16,160
cbba-tree	32,807	81,153	28,275	23,103
cbba-etc	31,800	27,853	4,338	23,091

Table 2: Summary of key performance metrics for the baseline scenario ( base_R20_V100 ). All values represent totals accumulated over 50 trials.

Figure 6: Mean performance of all evaluated algorithms in the baseline scenario, which consisted of 20 robots and 100 active victims over 50 trials. The bars represent the average number of victims rescued versus those lost per trial. The high number of ’Incompleted’ tasks reflects victims that expired (100s lifetime) in this challenging steady-state scenario where tasks are continuously replaced. Error bars indicate the 95% confidence interval. The BT-based consensus algorithms (‘cbba-tree‘, ‘cbba-etc‘) demonstrate significantly higher effectiveness.

Figure 7: Comparison of the mean communication overhead for each algorithm in the baseline scenario. The bars show the average number of messages sent per trial. The ’cbba-etc’ algorithm demonstrates an important reduction in communication compared to other consensus-based and communicative methods

Figure 8: Distribution of rescued victims per trial for each algorithm in the baseline scenario. The boxplot illustrates the median, quartiles, and range of performance over 50 trials. The results for ’cbba-etc’ and ’cbba-tree’ show similar effectiveness and are significantly higher than the other algorithms

Figure 9: Scalability of the algorithms with respect to task density, showing the percentage of rescued victims as the number of initial victims increases from 25 to 500, with a fixed team of 20 robots. The consensus-based algorithms that use Behavior Trees (’cbba-etc’ and ’cbba-tree’) maintain the highest performance up to moderate saturation (V200). At high saturation (V500), ’cbba’ and ’c-cbba’ achieve slightly higher absolute rescue counts, but Figure 10 reveals this comes at a significant communication cost.

Figure 10: Communication cost as victim density increases (logarithmic scale) for a fixed team of 20 robots. While ’cbba’ and ’comm’ exhibit exponential increases in messages, and ’c-cbba’ shows significant scaling, ’cbba-etc’ maintains high performance with only a marginal increase in communication, demonstrating superior efficiency at scale.

Figure 11: Scalability of the algorithms with respect to robot density, showing the percentage of rescued victims as the number of robots increases from 5 to 40, with a fixed set of 100 victims. The consensus-based algorithms that use Behavior Trees (’cbba-etc’ and ’cbba-tree’) maintain the highest performance across all swarm sizes, while ’c-cbba’ consistently performs in the lower tier.

Figure 12: Performance degradation of each algorithm as the probability of action failure increases from 0% to 50%. The lines plot the mean number of rescued victims, with shaded areas representing the 95% confidence interval. Critically, the non-BT architectures (’c-cbba’, ’cbba’) collapse at high failure rates, while all BT-based architectures demonstrate significant robustness.

Figure 13: System resilience to permanent agent failure, showing the mean number of rescued victims as the probability of agent loss per step increases. The performance of all decentralized algorithms degrades gracefully. ’cbba-etc’ and ’cbba-tree’ maintain their high-performance advantage even as the number of active robots in the swarm decreases over time.

Figure 14: Algorithm robustness to communication degradation, plotting the mean number of rescued victims against increasing packet loss probability (with bandwidth limits). The BT-based consensus architectures (’cbba-etc’ and ’cbba-tree’) consistently outperform other methods across all tested levels of communication failure, demonstrating significant resilience.

Figure 15: Analysis of heterogeneity management efficiency versus communication cost in the finite-task scenario. The plot shows Total Failed Rescues (due to capability mismatch) against Total Messages Sent (logarithmic scale). The ideal performance region is the bottom-left corner (minimal failed rescues and minimal communication). Cbba-etc and c-cbba demonstrate the best overall balance, achieving low communication cost while effectively managing heterogeneity, positioned closer to the ideal region compared to the communication-heavy cbba and comm algorithms. For visualization purposes on the logarithmic message axis, the non-communicating tree algorithm is not shown. It performed worst in managing heterogeneity, accumulating 7,607 failed rescues because it cannot share or receive information about task requirements.

Figure 16: A cost-benefit analysis of the algorithms under baseline conditions, plotting total rescued victims against the total number of messages sent (on a logarithmic scale). This visualization highlights the trade-off between mission effectiveness and communication efficiency. ’Cbba-etc’ is positioned in the upper-left, demonstrating its performance with minimal communication cost.

Multi-Robot Task Allocation

Multi-robot task allocation assigns work across robot teams while balancing capabilities, costs, timing, and coordination constraints.

See all 112 papers

FORMICA: Decision-Focused Learning for Communication-Free Multi-Robot Task Allocation
Feb 20, 2026Antonio Lopez, Jack Muirhead, Carlo PinciroliMulti-Robot Task Allocation Multi-Agent Resource Allocation
Dept. of Robotics Engineering, Worcester Polytechnic Institute, Worcester MA 01609, USA
Dynamic Multi-Robot Task Allocation under Uncertainty and Communication Constraints: A Game-Theoretic Approach
Apr 13, 2026Maria G. Mendoza, Pan-Yang Su, Bryce L. Ferguson +1Multi-Robot Task Allocation Multi-Agent Resource Allocation
Department of Mechanical Engineering, University of California, Berkeley, Berkeley, CA, United States · Department of Electrical Engineering and Computer Science, University of California, Berkeley, Berkeley, CA, United States · Thayer School of Engineering, Dartmouth College, Hanover, NH, United States
Auction-Consensus Algorithm with Learned Bidding Scheme for Multi-Robot Systems
May 21, 2026Jose Rodriguez, Constantine Tarawneh, Sven Koenig +2Multi-Robot Coordination Multi-Robot Task Allocation
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

Feb 20, 2026J/K move · Enter open · S save

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

Antonio Lopez, Jack Muirhead, Carlo Pinciroli

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

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

Open paper AskPDF arXiv

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.

Abstract

Multi-Robot Task Allocation

Explore similar work

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

Dynamic Multi-Robot Task Allocation under Uncertainty and Communication Constraints: A Game-Theoretic Approach

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