Sustainable Multi-Agent Crowdsourcing via Physics-Informed Bandits

Chayan Banerjee

cs.MAFeb 25, 2026

Sustainable Multi-Agent Crowdsourcing via Physics-Informed Bandits

Authors: Chayan Banerjee

Organizations: School of Electrical Engineering and Robotics, Queensland University of Technology (QUT), Brisbane, QLD, Australia.

Abstract

Crowdsourcing platforms face a four-way tension between allocation quality, workforce sustainability, operational feasibility, and strategic contractor behaviour--a dilemma we formalise as the Cold-Start, Burnout, Utilisation, and Strategic Agency Dilemma. Existing methods resolve at most two of these tensions simultaneously: greedy heuristics and multi-criteria decision making (MCDM) methods achieve Day-1 quality but cause catastrophic burnout, while bandit algorithms eliminate burnout only through operationally infeasible 100% workforce utilisation.To address this, we introduce FORGE, a physics-grounded $K+1$ multi-agent simulator in which each contractor is a rational agent that declares its own load-acceptance threshold based on its fatigue state, converting the standard passive Restless Multi-Armed Bandit (RMAB) into a genuine Stackelberg game. Operating within FORGE, we propose a Neural-Linear UCB allocator that fuses a Two-Tower embedding network with a Physics-Informed Covariance Prior derived from offline simulator interactions. The prior simultaneously warm-starts skill-cluster geometry and UCB exploration landscape, providing a geometry-aware belief state from episode 1 that measurably reduces cold-start regret.Over $T = 200$ cold-start episodes, the proposed method achieves the highest reward of all non-oracle methods ( $\text{LRew} = 0.555 \pm 0.041$ ) at only 7.6% workforce utilisation--a combination no conventional baseline achieves--while maintaining robustness to workforce turnover up to 50% and observation noise up to $σ= 0.20$ .

Figures & tables

Fig. 1: FORGE Simulator — $K{+}1$ Multi-Agent System. The offline phase pre-computes a Physics-Informed Prior from dataset $\mathcal{D}_{\text{sim}}$ , injecting initial weights $\boldsymbol{\theta}_{0}$ and covariance $\mathbf{A}_{0}^{-1}$ into the Allocator at $t=1$ . During live allocation, each contractor independently declares availability $a^{c}_{t,k}$ via a fatigue-threshold policy before the Allocator acts. Observable state variables and the availability signal are concatenated with the task query into a 493-dimensional context vector $\mathbf{x}_{t,k}$ , which drives the Allocator’s selection $a_{t}$ . The Marketplace Environment processes this decision through three parallel dynamics—success probability, fatigue (RMAB), and surge pricing—feeding reward $r_{t}$ back to the Allocator and updated states back into the context vector each episode.

Method	ERew	EReg	LRew	LReg	Burn	Util	SMR%	MPF
Greedy (Max-Rep)	$0.454{\pm}0.039$	$37.97{\pm}3.81$	$0.441{\pm}0.080$	$19.65{\pm}3.46$	$22.8{\pm}3.5$	$11.8{\pm}1.2$	$36.2{\pm}1.1$	$0.456{\pm}0.008$
Greedy (Min-Price)	$0.464{\pm}0.057$	$37.30{\pm}6.52$	$0.463{\pm}0.042$	$18.86{\pm}1.96$	$6.2{\pm}4.1$	$13.8{\pm}2.1$	$8.9{\pm}3.9$	$0.162{\pm}0.049$
TOPSIS	$0.534{\pm}0.036$	$30.32{\pm}3.51$	$0.538{\pm}0.086$	$15.11{\pm}4.27$	$28.8{\pm}3.7$	$5.8{\pm}0.7$	$38.4{\pm}2.4$	$0.485{\pm}0.016$
LinUCB	$0.442{\pm}0.015$	$39.48{\pm}1.43$	$0.518{\pm}0.026$	$16.10{\pm}1.48$	$0.2{\pm}0.4$	$99.6{\pm}0.5$	$0.4{\pm}0.5$	$0.023{\pm}0.006$
SW-UCB	$0.452{\pm}0.013$	$38.51{\pm}0.75$	$0.455{\pm}0.012$	$19.24{\pm}0.59$	$0.0{\pm}0.0$	$62.4{\pm}2.6$	$0.0{\pm}0.0$	$0.005{\pm}0.003$
Thompson Sampling	$0.449{\pm}0.023$	$39.00{\pm}3.07$	$0.449{\pm}0.022$	$19.12{\pm}0.87$	$\mathbf{0.0{\pm}0.0}$	$96.6{\pm}2.1$	$\mathbf{0.0{\pm}0.0}$	$0.006{\pm}0.003$

TABLE I: Comparative performance over

T{=}200

episodes with contractor agency enabled (mean

\pm

std, 5 seeds). ERew/EReg = Early Reward/Regret (Phase 1A,

t\!\in\![1,100]

); LRew/LReg = Late Reward/Regret (Phase 1B,

t\!\in\![150,200]

); Burn = Total Burnout Events; Util = Unique Contractors Used; SMR = Strategic Misrouting Rate (% episodes where selected contractor signalled

a^{c}\!=\!0.5

); MPF = Mean Pre-Selection Fatigue. Bold = best per column (excluding oracle).

Variant A: TOPSIS
Turnover / Noise $\sigma$	0.00	0.05	0.10	0.20
0%	59.5	59.5	59.5	59.5
10%	57.8	57.8	57.8	57.8
30%	61.2	61.2	61.2	61.2
50%	54.5	54.5	54.5	54.5
Variant B: Hybrid (No Prior)

TABLE II: Cumulative Regret under varying workforce turnover and observation noise with contractor agency enabled (10 repeats/cell,

T{=}200

). Lower is better. Bold = best per cell (excluding TOPSIS baseline).

TOPSIS (Baseline)
Surge ( $\omega$ ) / Noise ( $\sigma$ )	0.00	0.10	0.20
1.0 $\times$	28.80	28.80	28.80
1.5 $\times$	75.20	75.20	75.20
2.0 $\times$	97.40	97.40	97.40
Hybrid (No Prior)
1.0 $\times$	28.20	28.60	28.40

TABLE III: Burnout Events under systemic surge stress with contractor agency enabled (10 repeats/cell). Lower is better. Bold = best per cell (excluding TOPSIS baseline).

TOPSIS (Baseline)
Surge ( $\omega$ ) / Noise ( $\sigma$ )	0.00	0.10	0.20
1.0 $\times$	0.54	0.54	0.54
1.5 $\times$	0.41	0.41	0.41
2.0 $\times$	0.35	0.35	0.35
Hybrid (No Prior)
1.0 $\times$	0.54	0.54	0.54

TABLE IV: Mean Reward under systemic surge stress with contractor agency enabled (10 repeats/cell). Higher is better. Bold = best per cell (excluding TOPSIS baseline).

Explore similar work

From Novice to Expert: Cost-Aware Bandits for Evolving Worker Performance in Crowdsensing
Jul 15, 2026Yin Huang, Qingsong Liu, Jie XuMulti-Armed Bandits Non-Stationary Bandits
Manning College of Information and Computer Sciences, University of Massachusetts Amherst
Guiding Worker Self-Selection in Crowdsourcing Contests: An LLM-Augmented Algorithmic Approach
Sep 7, 2026Nguyen Thach, Hau Chan, David Parkes +1Automated Algorithm Discovery Game Theory
University of Nebraska-Lincoln Lincoln, Nebraska, USA · Harvard University Cambridge, Massachusetts, USA
Collaborating in Multi-Armed Bandits with Strategic Agents
May 13, 2026Idan Barnea, Ofir Schlisselberg, Yishay MansourMulti-Armed Bandits Mechanism Design
Tel Aviv University · Google Research

Jul 15, 2026J/K move · Enter open · S save

From Novice to Expert: Cost-Aware Bandits for Evolving Worker Performance in Crowdsensing

Yin Huang, Qingsong Liu, Jie Xu

Manning College of Information and Computer Sciences, University of Massachusetts Amherst

Multi-Armed Bandits Non-Stationary Bandits Online Resource Allocation

Open paper AskPDF arXiv

Mobile crowdsensing (MC) recruits mobile users to perform sensing tasks using their smartphones, enabling large-scale applications such as traffic monitoring and environmental sensing. A fundamental challenge is online worker recruitment under uncertainty, where the platform must learn workers' sensing performance while operating with a limited budget. Existing learning-based MC recruitment methods typically assume that each worker's sensing quality is stationary with a fixed mean over time. In practice, however, worker performance often improves with experience and eventually stabilizes, while the incurred sensing cost can be unknown in advance due to time-varying device and context states. In this paper, we study a budget-constrained online recruitment problem in which the platform selects one worker in each round, observes the sensing quality and incurred cost, where the expected sensing quality of each worker increases with experience and eventually converges to a plateau, and repeats until the budget is exhausted. We formulate this problem as a structured bandit model where each worker's expected reward evolves according to an unknown increasing-then-converging function of its participation count, and each worker has an unknown expected cost. We develop a cost-aware online learning framework that jointly learns evolving reward trajectories and heterogeneous costs, detects performance saturation, and allocates the limited budget to maximize long-term sensing utility. We provide theoretical performance guarantees and validate the proposed approach through extensive experiments, demonstrating consistent improvements over baselines that ignore experience-driven dynamics or assume known costs.

Abstract

Explore similar work

From Novice to Expert: Cost-Aware Bandits for Evolving Worker Performance in Crowdsensing

Guiding Worker Self-Selection in Crowdsourcing Contests: An LLM-Augmented Algorithmic Approach

Collaborating in Multi-Armed Bandits with Strategic Agents