Starting date: 2022
Project RAPID – Consortium: Nexter Robotics/MASA Group
OBJECTIVE
The main objective of the SERBERE project is to delegate the surveillance of areas to heterogeneous swarms of autonomous patrol robots, tireless nocturnal sentinels.
We propose to work on the following operational, scientific and technical axes:
• The analysis of the environment by the robot and the interaction with the latter (complexity of the terrain and data fusion).
• The decision-making autonomy of robots and the necessary security guarantees that result from it,
• Explainability and interpretability by humans of decisions made by robots,
• Cooperation between heterogeneous robots,
• Evaluation of the extensibility and flexibility of the system by adding new effectors or new missions.
This project aims to explore the use of a decision-making artificial intelligence technology, Direct AI, hitherto mainly used for the empowerment of agents in simulations, within the framework of existing patrol robots.
We propose a dual approach aimed at carrying out a reflection in two environments:
• A simulated environment making it easy and quick to assess the relevance of the decision-making models implemented as well as the mechanics of information exchange and collaboration between robots.
• A real environment allowing an initial simulation of the concepts and solutions retained within the framework of a prototype implementing robots within an evolving environment
SERBERE areas of study
Conceptual axis
- Issue of the explainability and interpretability of the decision
- Be able to guarantee the non-occurrence of undesirable behavior
- Evaluation of an adaptive hierarchical organization solution
Technical axis
- Creation of autonomous behaviors for patrol robots
- Implementation of a collaboration solution between robots with heterogeneous capacities
- Management of the coactivity with the man
Operational axis
Demonstration on a swarm of real patrol robots in a representative environment
- Implementation of solutions for evaluating the performance of the proposal
- Evaluation of the extensibility and flexibility of the solution (software, architecture, hardware), i.e. the ability to be easily adapted when changing hardware, effectors, missions, etc.
- Issue of the interpretability of the decision by an operator
Demonstrative video
In the presence of STAT (la Section Technique de l'Armée de Terre) and AID (l'Agence de l'innovation de défense), MASA and Nexter joined forces to deliver a demonstration in which an ULTRO made full use of a MASA designed brain - driven by the embedded Artificial Intelligence engine DAI RT (RealTime), a MASA product - by successfully completing a complex patrol mission in full autonomous mode. The mission required the ULTRO to detect and avoid fixed and moving obstacles, both human and machine, while intelligently managing automatic recharging, and the loss of both positional awareness and the ability to communicate. Further to these capabilities, the module is able to pursue an intruder, when requested to do so by the operator.
The SERBERE project, which is managed by AID, is therefore in the process of completing its first-phase objective for individual robots, and will now turn its attention to the second phase, in which multiple heterogeneous robots combine in swarms to complete complex missions. The resulting decisional model, will be transposable to other robots, and through reuse will be put forward as a fully-fledged swarm command model.
In case you were unaware, MASA DAI is an Artificial Intelligence SDK that provides for the incremental development of behaviors that are compromise-based, and easy to understand, maintain, and modify.
In the example referred to above, the robot receives a mission that requires it to patrol a clearly defined area, which involves choosing points of interest in a random fashion, while simultaneously managing events and alerts in an entirely autonomous manner.