4IOT robots: towards a less finite network and beyond

" We have been studying how robots can be used as mobile, intelligent elements of telecommunications network infrastructure," introduces David Gesbert, communications researcher and recently appointed Director of EURECOM. Cellular networks like 5G are organized via relay antennas that are placed by different operators on radio towers. These antennas ensure connectivity with users and objects on the ground. However, some areas such as rural environments or isolated locations lie outside the network's coverage range. Fixed network infrastructures can also be damaged by bad weather.

The Robots 4IOT project led by David Gesbert at EURECOM aims to extend and improve network coverage using a mobile radio relay approach. The team's strategy is to place relay antennas on flying drones that move autonomously and intelligently according to network coverage. These robots position themselves between the terrestrial radio towers and users or objects on the ground further away, to provide the connectivity service. The optimal positioning of the robots is calculated according to the environment and local radio wave propagation.

From algorithm to rhythm

But how do you adjust a robot's trajectory in real time in a new environment? The basic idea is for the robot to decide where to place itself, guided by the measurements it makes. Intelligent algorithms based on machine learning allow us to sense the direction of movement in which the robot will be able to move, and to gather the information that is most relevant to the task the robot is supposed to perform.

"Different approaches using different algorithm construction strategies are possible," explains David Gesbert. The team's preferred approach is called reinforcement learning. This involves providing the robot with a minimum of initial information and letting the algorithm learn from the measurements taken of the environment during the drone's trajectories. In this way, the robot learns the best possible strategy as the situation progresses. "This type of learning is particularly used in games, such as chess or go, but is not yet widely used in robotics, as it is associated with major technical challenges and risk-taking", points out the researcher.

Another, more traditional approach is to feed information into the algorithms, in this case three-dimensional maps of the environment and its obstacles. This reduces the time and energy invested in learning, but the robots are then less able to adapt to new environments or changes that may occur within them. "We favor reinforcement learning, but we also inject models that are sufficiently close to reality to reduce the number of hypotheses to be tested," explains David Gesbert. For the moment, the robot prototypes are being assembled by the team and trajectory optimization tests are being carried out on the EURECOM campus.

Mapping and sensing: research priorities for 6G

"In a second phase, the scope of applications for this type of robot could extend beyond telecommunications network optimization," continues David Gesbert. They could, for example, help map locations in detail using refined wave propagation models, where the power of waves reflected by objects on the ground reflects the various obstacles and their composition. While some 3D maps already exist, this approach would enhance their level of detail. "This would involve optimization work to take into account the fine features of the terrain", points out the researcher. In turn, the mapping data affecting the radio field can be used by the robot to pinpoint the precise location of a radiating object in its environment, such as a connected object or a user terminal.

In the long term, it is also conceivable that flying robots will feature different types of on-board sensors, such as radar, lidar or cameras, and that data fusion will further enhance the level of detail obtained about environments. These robots could also be used in the deployment of an Internet of Things (IoT) network, collecting different types of data from sensors. For example, along their trajectory, they could collect data supplied by sensors placed on industrial installations or in connected cities, such as pressure, humidity or temperature data. This type of application, known as sensing, is one of the fields of investigation for the sixth generation of mobile telecommunications.

This part of the project is currently being prototyped at EURECOM.

EURECOM: partnership research for the industry of the future

Founded in 1991 as a consortium of academic and industrial partners, EURECOM is an engineering school and research center in digital sciences with a strong international vocation. Institut Mines Télécom is a founding member of the consortium. EURECOM deploys its expertise in three main areas: data science, digital security and networks of the future.

Opening up new avenues towards the technologies of the future and emerging technologies, and fostering partnerships with companies to promote the transfer of knowledge towards the Industry of the Future are key strategic thrusts of EURECOM's policy.

EURECOM has held the Carnot label jointly with ITM since 2006, in recognition of the quality of its research partnerships.