Description
Date depot: 27 décembre 2021
Titre: Distributed Radar with Auxiliary Sensor Collaboration by Multi-Agents Programming.
Directrice de thèse:
Amal EL FALLAH SEGHROUCHNI (LIP6)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Intelligence artificielle
Resumé: Different applications in the field of consumer Services, Health and Security rely on the use of friendly drones
(or UAVs) whilst facing the threat of hostile drones flying over strategic infrastructures (Stadiums, Airports,
Energy and Defense facilities…).
Coupling a radar sensor for search with an auxiliary sensor (e.g. camera or radar) for confirmation, that will
increase drone detection relevance and lower false alarm rates, particularly for non-cooperative drones such
as those that are operating autonomously.
Drones have small radar cross-section, fly at low speeds in dense clutter environments, are highly
maneuverable and can be easily confused with natural (ex: birds) and manmade targets. This represents a
significant problem for their effective detection, recognition and tracking. This could generate high false tracks
rate disturbing the Command and Control (C2) system neutralization decisions efficiency. By coupling a Radar
with an auxiliary complementary sensor (e.g. camera or radar), we can drastically reduce false tracks by using
Radar capability to detect small flying objects at long ranges and in diverse weather conditions with auxiliary
sensor capability to discriminate drone from artefacts (clutter, birds, …). Capacity of focusing will be used for
target re-acquisition (assigned by radar), tracking and recognition.
When a drone track has been confirmed by the Radar, auxiliary sensor (e.g. camera or radar) resources will
be used to improve Radar tracking quality by reducing risks of drops (in case of non-detection), by improving
track accuracy (due to auxiliary sensor high update rate and angular resolution) and by maintaining tracks in
case of abrupt drone maneuver (drone stop/turn/acceleration). Auxiliary sensor resources for accurate target
tracking will be mobilized whenever the radar needs to insure continuity of tracks.
Auxiliary sensor (e.g. camera or radar)resources will then be used jointly to:
• Reduce false tracks at initiation phases
• Maintain quality of tracks when the drone threat has been confirmed.
Effective tracking aids robust recognition, yet in dense target environments, recognition features are
important for effective tracking.
In this Research study, we propose to rely on agent paradigm and to develop AI collaborative Multi-Agent
techniques to jointly solve the problem of recognition and tracking of drones, using sensors resources
optimization between a radar and an auxiliary sensor (e.g. camera or radar).
This work will be supported by real radar and auxiliary sensor measurements.
The goals addressed in this thesis are the following:
- Propose an architecture based on multi-agents allowing collaboration between a Radar and an
auxiliary sensor (e.g. camera orradar) considered by the Command a Control System (C2) as a MetaSensor
- Define a process for dynamic resources allocations between sensors to improve the quality of tracks
and to lower the rate of false tracks per day.
The agent paradigm is particularly recommended as a suitable receptacle with a modular and distributed
architecture [2]. At the agent level, Radar and auxiliary sensor can collaborate and produce a coherent
behavior to achieve a joint mission. A multiagent system paradigm [9] offers many tools for cooperation,
coordination and collaborative problem solving [12, 2] and could be enough generic to allow new sensors plug
and play without compromising the initial architecture.
AI background:
Suitable AI techniques for collaborative Radar/Auxiliary Sensor (e.g. camera or radar) Resources
Management could be integrated into Multi-Agents:
• Reactive agents/Cognitive agents/Hybrid agents
• MCTS-based Automated Negotiation Agents
• Distributed-Auctions Based Collaboration Agents
• Multi-Agents Reinforcement Learning
• Distributed Constrained Programming
Different kinds of architecture will be explored:
• Centralized Architecture
• Decentralized Architecture
• Hybrid Architecture coupling hierarchical decision processes and distributed ones
Thesis topics to be addressed include:
• Topic 1: Multi-agent architectures and modelling: Multiple agents act as sensors (Radar and Auxiliary
sensor) and uses suitable cooperation mechanisms such as distributed decision strategies for both
reduction of false tracks / day, and for improving track quality on confirmed threat of drones. Radar
will initiate detections that will be confirmed by auxiliary sensor (e.g. camera or radar) including
auxiliary sensor tracker and recognition capacity. When the target will be confirmed as a drone threat,
auxiliary sensor resources will assist radar tracker to maintain a high quality of tracking, by
reacquisition in case of track drop risk, and by high update measurement to help radar tracker to
converge to a minimal angular accuracy.
Doctorant.e: De Rochechouart Maxence