This project implements an innovative concept of coordination of robot teams based on anticipation networks (http://anticipatory-networks.pbf.biz/en). Within the industrial research phase of the project, a decision engine was developed to be embedded in the robots' navigation software system and ensuring their optimal coordination. Coordination software based on the principles of anticipatory networks provides higher harvesting efficiency by autonomous robots and better team collaboration and human-robot interaction than the baseline harvesting mode by individual (non-coordinated) robots. Moreover, the use of anticipatory networks makes it possible to replace direct exchange in the situation of communication disruptions or insufficient bandwidth of communication channels with anticipated results of robots operation according to known optimization algorithms.

The first stage of the project (implementation in 2022) comprised simulation studies of a team of robots with anticipatory coordinating software. The simulation showed the usefulness of anticipatory coordination for picking berries in real growing conditions. The real-life data were received from on a strawberry plantation in southern Poland. The innovative autonomous architecture of robots is based on

• the anticipatory principles of coordination and control, and on
• the fusion of heterogeneous information from the sensor system, and P2P exchange in a simulation model and in a common knowledge base.

The software also integrates the exchange of information about the working environment made available in the Agricultural Internet of Things (AIoT). The main innovative elements of the entire software system of the robot team are:

- the concept of anticipatory coordination of a team of fruit-picking robots

- overall software architecture to transfer simulation procedures to control and coordinating software and

- optimization of harvesting parameters based on the use of the world model (WoMo) and cooperation with SLAM (Simultaneous Localization and Mapping) software.

While the robots are working, WoMo is constantly updated based on data collected from the environment using the sensors of each robot. The data are transferred to a common knowledge base of all robots. The fruit plantation simulation shows that this mode of operation requires efficient communication with the infrastructure of the working environment forming the AIoT ecosystem, as well as with operators and external knowledge bases. In addition, WoMo constantly exchanges information with the autonomous decision engines of each individual robot. This enables robots to share relevant information about their working environment. One of the robots (the current coordinating unit) uses the information received from SLAM to update the base of rules used in the collection algorithms and the map of the environment. All of the above elements are used to plan the harvest target, start picking fruit at the specified destination by each robot, and perform other activities required for efficient harvesting.