Adaptive Autonomous Production Platform for Environment – AdAP2E

In order to tackle the variability of situations, robust perception and control approaches are implemented to maintain the performance of the robot regardless of the environment encountered. In particular, adaptive principles help to feed hybrid models of representation, able to take into account the uncertain dynamics. In addition, these models are operated to modulate the actions of the robot and to ensure its stability and performance in terms of accuracy.
Given the diversity of situations in the field-road, these moderations in control signals may lead to failure situations: the robot is physically unable to continue its evolution with the required constraints. As a result the mechanical, reconfiguration is mandatory in order to keep on working. While hybrid models are used to estimate the capability of the robot, they can be exploited to determine a mechanical configuration allowing the continuation of the mission. The project thus exploits the concept of mechatronics by determining through control laws, the degrees of freedom required to adapt the mechanical structure. To do this, innovative design methodologies are implemented to design a reconfigurable chassis capable of offering behavioral variability compatible with the diversity of actions and conditions robot / environment interaction.
Nevertheless, several possible configurations and several approaches to control, requiring multiple sensors are required for changes in natural environment (monitoring structure, target tracking). Also, concurrent different operating modes must be defined and selected and weighted in real time. Also, decision-making based especially on fuzzy logic and Markov chain are considered in this project.

To take into account the variability of the conditions for interaction between the robot system and its environment, several approaches for perception and control have been developed based on typical scenarios encountered in environmental tasks. These behaviors have been tested and qualified through full scale experiment. In particular 5 modes were implemented: target tracking by laser, structure following by vision-lidar fusion, trajectory tracking using GPS, evaluation of traversability by 3D reconstruction, obstacle crossing by force feedback. These first modes of travel afforded original contributions, especially in the field of adaptive control at different speeds (and therefore with very varying dynamics), maintaining a unique architecture of control. If each of these modes can achieved specific contribution, the main target lies in the selection of these basic behaviors. Indeed, from an algorithmic perspective, the goal is now to implement decision-making processes, based on the recognition of the context of the task, and on-line task planning.
Innovative design methodologies have permitted to generate concepts of reconfigurable chassis that meet the specifications, but still out of the financial framework of the project. Nevertheless, a concept capable of combining modularity within financial constraints is the implementation of a multi-robot system, generalizing the concept advanced by the project and generating new modes of perception and control. Today the realization of 2 joinable robots is being started, and algorithmic developments have focused on the synchronization of several robots, increasing the interest of the supervision process.

The results related to the modular architecture of the robotic system have led to the development of a multi-robots architecture of control capable of providing a high level of modularity. This opens a way to an extension on the topic of mechanical association of generic robotic entities. In fact, this should lead to the proposal of a european project H2020 related to cooperating robot in the field of agriculture. Indeed, the first results on synchronizing two robots suggests also the robot cooperation between different types (in particular between ground and flying robots).
The scope has openned the way for several related projects in connection with industrial, for the development of autonomous agricultural machinery. Beyond the robotic aspects, the possibilities offered by the supervisory approaches allow to propose short-term advances on aid to the decision making and big data collection.
Finally, the integrity preservation of mobile robots can be applied directly on the development of safety devices for off-road vehicles. Several cooperations with companies have been initiated in this direction.

Two main scientific contributions on the topic of adaptive and predictive control has been achieved in the framework of Adap2E project, in high level journals :
Roland Lenain, Jean-Baptiste Braconnier, Benoit Thuilot, Vincent Rousseau, Robust sideslip angles observer for accurate off-road path tracking control, Advanced Robotics, accepté.
Audrey Guillet, Roland Lenain, Benoit Thuilot, Vincent Rousseau, Formation control of agricultural mobile robots: a bi-directional weighted constraints approach, Journal of Field Robotics, accepté

A communication in a conference, on the applicative aspects has also been achieved :
Bernard Benet, Vincent Rousseau, Roland Lenain, Fusion between a color camera and a TOF camera to improve traversability of agricultural vehicles, Ageng 2016, Aarhus, Danemark

Several papers in conference are also under submission. Beyond scientific publications, the patentability of the multi-robots approach for achievement of agricultural task is under investigation.
Finally, several exhibition have been achieved with already available small robots :
- Paris International Agricultural Show 2016
- Europa day 2016, at the european comission

This project takes place in the design of innovative robotic tools to improve tasks to be achieved in the field of environment and agriculture. It aims at increasing the efficiency of operations to be performed off-road, while improving the environment preservation. The achievement of autonomous robots in this context faces several points, to be tackled with a multidisciplinary approach : algorithm decision support , human/machine interaction, robotics control and perception. It proposes to develop an interactive and reconfigurable robotic demonstrator able to perform accurate motion in natural environments, with an adjustable degree of autonomy a, with respect to the encountered situation and the task to be done. This research group is expecting to demonstrate the capabilities of its 42 months woks thanks to an actual robot, moving in full-scale conditions.

Based on the mobile robots control theory, the AdaP2E project take place on the synthesis of several motion control approaches, such as path following, moving object tracking, high speed motion or traversability management. Some of these elementary behaviors are already available, in particular accounting for off-road specificities (such as grip conditions, or uneven ground). Nonetheless, these approaches uses different kind of sensors, are adapted to different kind of robots and dedicated to some framework. As a result, their performances are related to the situations, and none of the control laws may be applied in all the conditions encountered when achieving off-road operations. Furthermore, a preliminary planning is mandatory, when considering a high autonomy level.

Therefore, the team arising with this project is focused on the design of a high level software structure, allowing to manage the selection of elementary control modes and their stable transition at several levels. First, the task planning is considered with respect to the software (perception and control algorithm) and the hardware (sensors available, robot configuration). If planning may be viewed as an off-line process, the on-line monitoring is considered as a second level, allowing to manage on-line the different control laws in order to adapt the robot behavior to the environment, the task to be achieved, and people in interaction with the system. The team will then attach great importance to the consideration of the human dimension. Indeed, it will guide the development of supervision to take into account the expertise of the operators with respect to reference profiles .

Researches related to the control, supervision and man-machine interaction may bring scientific contributions by themselves, particularly when considering off-road specificities (in view of the large parametric variability). However, the major contribution is undeniably part of the integration of these different areas within a global and interactive architecture. This is made possible by the establishment of a research team in the theme of reconfigurable mobile robotic systems. The main research interest will be focused on the on-line adaptation in order to tackle variable and uncertain phenomena encountered off-road. The objective is to make robotic systems able to cope with different situations in order to meet operators expectations and help them with work in natural environment. These operations indeed appear to be difficult and dangerous, whatever the field of application of the proposed project.