Plant stresses and diseases cause major losses to agricultural productivity and quality. Left unchecked, stresses and diseases can spread and propagate to nearby plants, causing even more damage, necessitating early detection. To address this challenge, the Agricultural Robotic System for Plant Stress Propagation Detection (ARS/PSPD) is developed. In this cyber-physical system, the robot agents are assigned scanning tasks to detect stresses in greenhouse plants. The problem of plant stress propagation detection is formulated with disruption propagation network modelling, which captures the plant stress occurrence and propagation mechanisms. The network modelling enables better situation awareness and augments the development of advanced collaborative scanning protocols. Five collaborative scanning protocols are designed and implemented in this research, with one protocol serving as a baseline, three protocols utilising disruption propagation network analysis, and one protocol utilising Bayesian network inference. The scanning protocols minimise errors and conflicts in scanning task allocation and enable better plant stress detection. The five ARS/PSPD collaborative scanning protocols are validated with numerical experiments, using agricultural greenhouses as experiment settings. The experiments show that the scanning protocol using Bayesian network inference outperforms all other protocols in all scenarios, with 16.92% fewer undetected plant stresses and 12.28% fewer redundant scans.