INTERNSHIP CROSSING 2023/24

CROSSING invites applications from undergraduate and Masters level students. CROSSING internships are based around multidisciplinary topics in human-autonomous agent teaming. A list of topics can be found below.

If you are interested, please contact the supervisors to discuss the internship details.

Each internship will take place over a minimum of 4 months and the student will receive a bursary of $5 000 AUD paid through reimbursement of invoices for international students and casual contract for Australian students.  Students applying from France or interstate in Australia will be eligible for travel support up to an additional $2 500 AUD. We aim to support a minimum of 9 interns in this call.
– Decison round 1 : October 20, 2023.
– Decision round 2: November 24, 2023.

 

#1- Development of an experimental platform to investigate distributed human-AI teaming

  • Supervisors: Prof Anna Ma-Wyatt (UoA) and Dr Melanie McGrath (Collaborative Intelligence Future Science Platform/Responsible Innovation Future Science Platform, CSIRO)
  • Contact : anna.mawyatt@adelaide.edu.au, melanie.mcgrath@data61.csiro.au
  • Summary: In an operational environment, people often work as part of a distributed team. Team performance should also be resilient to changes in tempo and urgency. The hope is that a human-AI team will provide an opportunity to develop even more resilient teaming. However, as yet there are few experimental platforms to test human-AI teaming including the integration of AI algorithms. In this internship, you will work on adapting an experimental paradigm based on collaborative foraging. This experimental paradigm will enable the collection and analysis of data about human-AI teams (with real or simulated AI agents) while engaged in distributed, collaborative work. Within this paradigm, it will be possible to change the tempo and urgency of tasks. This project is particularly suited to students with a background in psychology, software engineering, or computer science and with an interest in teaming and human-AI interaction.

#3- Collecting pilot feedbacks during training with multiple autonomous drones in a simulated environment

  • Supervisors: Dr. Andrew Cunningham (UNISA/IVE), Dr. Linda Grosser (UNISA/BBB), Dr. Jean-Philippe Diguet (CNRS)
  • Contact: andrew.cunningham@unisa.edu.au, jean-philippe.diguet@cnrs.fr
  • Summary: The first step of the project is the design of different scenarios using the available environment based on the Microsoft  AirSim/Unreal simulation platform. These scenarios will allow to test different behaviours of the autonomous drones (eg. manage with more or less accuracy to maintain a formation around the drone controlled by a human) and to rate the performance of the human operator controlling the mission (eg. accuracy of the trajectory during a patrol). The second step is the design and test of different solutions to capture at runtime human feedbacks to feed a Multi-Agents Reinforcement Learning. This work will be done with VR/AR and Human factors experts of CROSSING. The objective is the design of a simple and user friendly way for the human to operator to regularly rate the autonomous drones during the mission. Different options will be considered starting with conventional interfaces such as mouse, keyboard or joystick that will provide a baseline. Then new interfaces that are expected to be more performant in VR than traditional desktop will be explored.

#4- Modelling and simulation of realistic faults in an unmanned underwater vehicle

  • Supervisors: Katell Lagattu (PhD Student, ENSTA/FLINDERS/NG), Prof. Benoit Clement (ENSTA/CNRS), Dr. Eva Artusi (Naval Group)
  • Contact: katell.lagattu@ensta-bretagne.org, benoit.clement@ensta-bretagne.fr
  • Summary: Unmanned Underwater Vehicles (UUVs) have emerged as a prominent and accepted solution, particularly in the naval defence sector. A significant challenge in UUV mission’s is ensuring a safe and efficient control despite the occurrence of faults, which refer to an unpermitted condition that changes at least one characteristic property of a component. Faults during a mission are inevitable and can lead to a drop in performance, the termination of the mission or even the loss of the drone. This is why it is necessary to take faults into account when designing the drone’s control system. In order to evaluate the control performance realistically, these faults need to be modelled and simulated. The project deals with faults caused by external factors, such as collisions. The focus will be on deformations of the drone’s hull, and on partial and total faults of the actuators: propellers blocked or entangled by a foreign object, damage to the propellers, damage to the fins, erosion, etc.

#5- Application of Time-Series Methods to Predict Physiological Data

  • Supervisors: Dr. Ehsan Abbasnejad (UoA/AIML), Dr. Jean-Philippe Diguet (CNRS).
  • Contact: ehsan.abbasnejad@adelaide.edu.au, jean-philippe.diguet@cnrs.fr
  • Summary: CROSSING has collected physiological data (Eye Tracker, Heart beat, Brain activity, Skin conductance, …) from about 20 participants executing different of psychology tasks for about an hour. These data have been used to train Machine Learning methods to predict human performance (success / failure rate). Our preliminary results are promising, the objective of this project is to explore different time-series approach such as to predict the evolution of physiological data to extend the prediction of performance in a near future.

#6- International Regulations for Preventing Collisions at Sea (COLREG) : Simulation

  • Supervisors: Prof. Benoit Clément (ENSTA/CNRS)
  • Contact: benoit.clement@ensta-bretagne.fr
  • Summary: Autonomous vessels have emerged as a prominent and accepted solution, particularly in the naval defence sector. However, achieving full autonomy for marine vessels demands the development of robust and reliable control and guidance systems that can handle various encounters with manned and unmanned vessels while operating effectively under diverse weather and sea conditions. A significant challenge in this pursuit is ensuring the autonomous vessels’ compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). These regulations present a formidable hurdle for the human-level understanding by an autonomous system as they were originally designed from common navigation practices created since the mid-19th century. Their ambiguous language assumes experienced sailors’ interpretation and execution and, therefore, demands a high-level (cognitive) understanding of language and agent intentions that are beyond the capabilities of current state-of-the-art of intelligent system.The intern will be in charge of developing new parts of an existing Python simulator in order to perform rapid assessment of a COLREGs situation with multiple vessels situation. Some complex tools exist but as the simulator will be dedicated to Learning algorithms we nee a very simple but realistic one. The second objective is to compare procedures (AI-based and others) that are COLREGs policy compliant.

#7- AR study for controlling an autonomous system

  • Supervisors: Dr James Walsh (UniSA/IVE), Dr. Étienne Peillard (IMTA)
  • Contact: James.Walsh@unisa.edu.au, etienne.peillard@imt-atlantique.fr
  • Summary: Previous work has shown that using augmented reality to display information related to the decision context of an autonomous system increases the user’s confidence in it. This was demonstrated in the case of a driving simulator, even when the information displayed was not related to the driving task. However, it is also known that the link between the user and the system can be enhanced by allowing the user to customize the system. The hypothesis that then emerges is that customizing the interface in AR could enable the user to increase trust.  This internship proposes to explore this question through an experiment. The work to be carried out will be i) complete a review of the state of the art concerning the link between customization, AR and trust, ii) propose an experiment to test the research hypotheses, iii) adapt the existing experiment simulator to run the experiment, iv) run the experiment and analyze the results with the supervising team and v) participate to the writing of a related research paper

#8- Simulation of multi Underwater Unmanned Vehicles in defense scenario including operators

  • Supervisors: Prof. Cedric Buche (Naval Group),
  • Contact: cedric.buche@pacific.naval-group.com
  • Summary: Development of the simulator that will be a tool to simulate multi-drones in underwater conditions, in a defense scenario including operators. Tool to be used and shared by NG PhD students. Technically based on Plankton, ROS2 and multiagents technics.

#9- Experimental study of perception in Virtual Reality

  • Supervisors: Prof. Anna Ma-Wyatt (UoA), Dr. Étienne Peillard (IMTA), Prof. Gilles Coppin (IMTA)
  • Contact: etienne.peillard@imt-atlantique.fr, anna.mawyatt@adelaide.edu.au
  • Summary: Augmented Reality (AR) allows to present digital images, objects and environments superimposed on the real world. However, due to technical limitations, the rendering of these digital elements remains different from the real elements. This internship proposes to study the specific case of orientation in virtual environments and its impact on distance perception. Previous studies have shown that users’ perception of distances is compressed for objects placed on the sides of an environment. They suggest that this is due to the position of the user, seated and oriented in one direction. The aim of this internship is to test this hypothesis by means of a user study, investigating how this effect varies according to the user’s posture and the type of virtual environment.This internship proposes to explore this question through an experiment. The work to be carried out will be i) complete a review of the state of the art concerning the link between AR perception of distances and user position, ii) propose an experiment to test the research hypotheses, iii) adapt the existing experiment simulator to run the experiment, iv) run the experiment and analyze the results with the supervising team and v) participate in writing the related research paper

#10- Learning significant characteristics in human motion to identify individuals

  • Supervisors: Dr. Mai Queyen Pham (IMT Atlantique / CNRS)
  • Contact: mai-quyen.pham@imt-atlantique.fr
  • Summary: To understand and characterize an individual’s movement is an important problem in computer vision and graphics and one that is still a significant challenge. Models that can effectively characterize an individual’s movement would be relevant to many domains, including health and medical applications (e.g. to describe or detect disease progression) or for industrial applications (e.g. human-robot interaction).This internship is a part of a project on realistic synthesis of human movement for individuals that includes two main goals. The first purpose of developing a model that learns the significant characteristics in body movement in order to identify an individual. The second purpose is to then use learnt characteristics of body movement to synthesize new movements that are presentative of this individual. This internship will tackle the first purpose. For instance, many works have investigated the analysis of the motion and movement of a particular type, especially in the context of gait recognition or general human activities. This internship is mainly based on two articles proposed to study gait recognition and analyzing human activities to identify a person. Their datasets, as well as their source code, are available.

#11- Ethics and decision aid

  • Supervisors: Dr. Rachel Stephens (UoA), Vincent Bébien (PhD Candidate, UoA/IMTA), Prof. Anna Ma-Wyatt (UoA) 
  • Contact: vincent.bebien@imt-atlantique.fr, rachel.stephens@adelaide.edu.au
  • Summary: Human decision is a process that has been described according to various models, addressing economical and expected utility points of views as well as more cognitive ones, where the bounded rationality of decision makers sets a more ecological framework. Amongst these models, one can refer to HACT (Bruni 2007) which decomposes into different stages of information processing and supports multiple combination of these different stages between the decision maker and an intelligent decision assistant. The objective of the internship is to study the way these different stages but as well the different levels of autonomy of the decision aid could be constraint or guided by an ethical framework, and to propose remediations to ethical biases in implementing the scheme of authority sharing. The analysis will rely on a concrete use case of nurses planning optimisation, and use some real data collected on site to be completed by simulated data.

#12- Interfacing multi-agent reinforcement learning (MARL) a with humans in a virtual world

  • Supervisors: Dr. Jean-Philippe Diguet (CNRS), Prof. Amer Baghdadi (IMTA), Lucas Fentanes Machado (PhD Candidate, UoA/IMTA)
  • Contact: jean-philippe.diguet@cnrs.fr, amer.baghdadi@imt-atlantique.fr
  • Summary: The objective is to interface a MARL algorithm with a photo-realistic environment in order to improve the continual training of a hybrid team (artificial and human agents). In this project we consider an homogeneous set of UAVs and we use the Microsoft AirSim/Unreal simulation platform based on the Unreal Engine. It provides the environment for the training of a human pilot controlling a Drones with autonomous drones around. Drones positions are extracted and can be used by a MARL algorithm to train the autonomous drones.  The project includes the 3 following tasks: i) Implementation of partially distributed MARL method adapted from previous work. This work will be done in close collaboration with a PhD student. ii) High-Speed training with different modes (eg. autonomous drones keep a given formation around the leader, self-exploration around the leader, etc.). iii) Training with Human in the loop based on trained models from 2). The collection of human feedbacks will be developed separately in another project.

#13- The role of collaborative social signals in trust & team processes within human-autonomous agent teams

  • Supervisors: Prof. Anna Ma-Wyatt (UoA),  Dr. Melanie McGrath, (CINTEL/CSIRO)
  • Contact: anna.mawyatt@adelaide.edu.au, melanie.mcgrath@data61.csiro.au
  • Summary: Among humans, non-verbal social signals contribute to cooperation and collaboration by conveying information, coordinating activities and communicating intentions. These social cues may include eye contact, gestures, facial expressions, body, language, proximity, tone of voice, and touch. By helping individuals understand another’s intentions, emotions, and attitudes, these collaborative social cues can enhance trust and support effective team performance. As we continue to develop the capability for human-AI teaming, it is becoming necessary to a) understand how different types of collaborative social signals contribute to teamwork processes and trust, b) identify methodologies for sensing these social signals in humans, and c) find ways to instantiate this information in human-AI teams.
    This project will investigate relationships between collaborative social signals, team processes, and trust in the human teaming literature and explore ways that this knowledge may be adapted to the human-autonomous agents teaming context. It is particularly suited to students with a background in human factors or psychology, strong skills in conducting literature reviews, and an interest in teaming behaviour.

#14- International Regulations for Preventing Collisions at Sea (COLREG) : Data Collection

  • Supervisors: Prof. Benoit Clément (ENSTA/CNRS)
  • Contact: benoit.clement@ensta-bretagne.fr
  • Summary: Autonomous vessels have emerged as a prominent and accepted solution, particularly in the naval defence sector. However, achieving full autonomy for marine vessels demands the development of robust and reliable control and guidance systems that can handle various encounters with manned and unmanned vessels while operating effectively under diverse weather and sea conditions. A significant challenge in this pursuit is ensuring the autonomous vessels’ compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). These regulations present a formidable hurdle for the human-level understanding by an autonomous system as they were originally designed from common navigation practices created since the mid-19th century. Their ambiguous language assumes experienced sailors’ interpretation and execution and, therefore, demands a high-level (cognitive) understanding of language and agent intentions that are beyond the capabilities of current state-of-the-art of intelligent system. The intern will develop a tool to import, visualize and exploit historical AIS (Automatic Identification System) relays of vessel movement information behaviour, e.g. position, heading, speed and vessel type to serve as a synthetic form of navigation experience.  By examining the AIS history, one can get an idea of the past behaviour of the vessel.

#15- Hardware In the loop Simulation of object tracking based on combined Event Based and Standard Cameras for UAV applications

  • Supervisors: Dr. Panagiotis Papadakis (IMTA), Dr. Jean-Philippe Diguet (CNRS)
  • Contact: panagiotis.papadakis@imt-atlantique.fr, jean-philippe.diguet@cnrs.fr
  • Summary: The internship takes place in the project LEASARD funded by CominLabs which aims to increase the navigation autonomy of UAV in search and rescue scenario while optimizing the energy efficiency of embedded systems for computer vision. In this project we aim to combine the LEASARD efficient implementation of object detection with a realistic simulator allowing experiments involving human operators and autonomous system including co-training. The internship will be dedicated more specifically to the implementation on a NVIDIA Jetson AGC Orin of a new object tracking algorithm that combines data flows from a standard embedded and an event-based camera to optimize the tradeoff between accuracy and energy efficiency. The GPU-based implementation will be tested and used in a simulation environment (AIRSIM[2] based on Unreal) that produces the video flows capture by a UAV in a photo-realistic (eg. Modular Neighborhood from EPIC game) already used by the team to run experiments with both autonomous UAV and human pilots. i) Implementation of standard yolov5 on the ORIN board, ii) Interface ORIN board/AIRSIM, iii) Implementation yolov5 for event-based detection, iv)  Test of a combination of the 2 methods and v) Participate in writing the related research paper.