This paper considers a group of autonomous robots performing collective tasks in an antagonistic environment. The environment can significantly hamper information exchange and decision-making in the group. Well-known methods are intended for performing collective tasks by a large group of relatively simple objects and have obvious advantages (high reliability, potential complexity increase, etc.). However, these methods neglect the specifics of collective behavior in an antagonistic environment. Moreover, they are inapplicable when a collective mission is fulfilled by a few objects expensive for production and maintenance. We propose a new approach to collective mission fulfillment based on the principle of collectivism and the Internet-of-Things paradigm. The novelty of the approach is that each robot in the group is valuable: with different resources and functionality, they assist each other. The robots execute different roles within the group, operating independently to fulfill a collective mission. The robots are equipped with observation sensors and motion detectors. The functions and resources of robots are represented as external services. A service-oriented mutual assistance architecture is designed for robots to resolve conflicts and problem situations caused by the environment: each robot can use the resources and functionality of other robots via online requests when needed. Such spontaneous relations provide a fundamental opportunity for the collective management of unpredicted situations and the possibility of changing behavioral scenarios when fulfilling a collective mission, leading to self-organization. The effectiveness of this approach is illustrated by an example: a group of robots attacks a complex target secured by a defense system.