Forsvarets Forskningsinstitutt (FFI) develops autonomy for unmanned vehicles in the air, land and sea. Autonomous vehicles can be more robust, solve tasks more efficiently, and be easier to control than existing solutions. FFI works in close collaboration with the Norwegian Armed Forces (Forsvaret) to look at operational concepts for future autonomous platforms.
Taking your master's thesis in collaboration with FFI requires that you are eligable for a Norwegian security clearance, and have a grade average at 3.5 or above on your university courses (calculator). Contact Tønnes Nygaard (tonnesfn@uio.no / office 415) if you have any questions about this.
Topic suggestions
Autonomous navigation by dirt road recognition
Military operations using autonomous Uncrewed Ground Systems (UGVs) in unstructured environments heavily rely on precise self-localization and good understanding of the vehicle’s surroundings. However, GNSS solutions might not be trusted for military applications, and we want to investigate solutions for GNSS-denied navigation.
For tasks such as driving along dirt roads, a UGV should be able to recognize the road based on its own perception sensors. We can also assume that the layout of the road network is known, for example from aerial photos or maps. For this project, we want to explore how to achieve autonomous navigation on dirt road networks for a UGV. This project contains two main parts:
- Global navigation: use perception for determining position in road network.
- Local navigation: use perception system for determining where to drive, both for maintaining position on the road and making correct decisions at intersections.
This project could be suitable for collaboration, where one focuses on global and one on local navigation.
Contact:
Marius Thoresen
marius.thoresen@ffi.no
97695523
Monocular terrain mapping for autonomous navigation
Autonomous Uncrewed Ground Systems (UGVs) navigating in terrain and in unstructured environments rely on perceiving their surroundings to know where to drive. Mapping of the immediate surroundings can be done using LIDAR, but LIDAR is an active sensor that easily can be detected. An alternative to using LIDAR is to use monocular cameras and estimating depth, for example using neural networks.
In this project, the task is to investigate if terrain mapping based on depth estimation can be a viable alternative to using LIDAR for autonomous driving in terrain.
Contact:
Marius Thoresen
marius.thoresen@ffi.no
97695523
Camera Based Pose Estimation
Military operations using autonomous Uncrewed Ground Systems (UGVs) in unstructured environments heavily rely on precise self-localization and good understanding of the vehicle’s surroundings. This is especially important when operating in swarming operations where cooperation and coordination with other agents is crucial. Typically, the localization of ourselves and other agents can be solved by using satellite-based localization systems such as Global Navigation Satellite System (GNSS). However, in the electromagnetic contested environments of modern warfare, these systems cannot be relied upon.
A promising and proven solution to the self-localization problem is to estimate this by coupling camera and inertial measurements. A myriad of methods and applications to solve this problem exists today. To complement these methods, FFI wants to look at image-based solutions to estimate the pose of other agents in its surroundings. An example of a solution would be to estimate the pose of a vehicle by pairing images with a vehicle description (e.g., a CAD-file).
Some main aspects of the assignment could be:
- A study of relevant literature exploring both the state of the art of existing systems and the possibility of improvements.
- A comparison of existing methods.
- A design of a framework/architecture addressing the problem described above.
- Implementing a lab test setup to verify the framework.
- Documenting the results in a thesis.
Contact:
David Kolden
david.kolden@ffi.no
90915641
Exploring the Collaborative Autonomy Tasking Layer
Collaborative Autonomy Tasking Layer (CATL) is a NATO specification of a toolset/framework used to “... enabling interoperable operation of unmanned systems in dynamic constrained environments, such as those with limited communications.” FFI wants to explore the use of this toolset on a swarm1 of robots to get a comprehension of the system’s strengths, weaknesses and possible enhancements. Important properties to explore are how CATL handles communication loss and different aspects of robustness in general. The candidate is free to define their own scenarios.
Some main aspects of the assignment could be:
- A study of relevant literature exploring the state of the art of frameworks and methods for collaborating systems.
- A comparison of existing methods/systems.
- Implementing a lab swarm test setup to benchmark the properties of CATL.
- Perform experiments by using the swarm with CATL.
- Documenting the results in a thesis.
Contact:
David Kolden
david.kolden@ffi.no
90915641
Intention aware behaviors for cooperating autonomous agents
For a swarm of agents working on different tasks, synchronization and coordination must be achieved to fulfill tasks correctly and efficiently. An effective way for an agent to coordinate with other agents is to keep track of their intentions and consider these when handling their own tasks to minimize interference or help others.
FFI wants to explore this domain in relation to cooperation between autonomous vehicles. Important aspects to investigate are methods for obtaining other agents’ intentions (e.g., communicating them directly or inferring them by observing other agents’ behavior) and how to handle these intentions in an interoperability setting. The candidate is free to define their own scenarios.
Some main aspects of the assignment could be:
- A study of relevant literature exploring the state of the art of methods and approaches.
- Comparison of existing methods.
- Implementing a lab swarm test setup that can be used to test and benchmark different methods.
- Perform experiments with the implemented swarm.
- Documenting the results in a thesis.
Contact:
David Kolden
david.kolden@ffi.no
90915641
Sensor fusion – LiDAR and camera
The combination of Light Detection and Ranging (LiDAR) and camera(s) are a common setup on Uncrewed Ground Vehicles (UGVs). These are typically used for mapping and classifying the surrounding environment, tracking objects of interest, and navigation.
FFI is interested in exploring new methods related to the tasks above using the combination of LiDAR and camera. Overall, the candidate is free to define the problem and direction. Examples of domains to investigate could be Simultaneous Localization and Mapping (SLAM), Object detection and classification, object pose estimation, calibration methods, etc.
Some main aspects of the assignment could be:
- A study of relevant literature exploring the state of the art of existing methods.
- A comparison of existing methods.
- A design of a framework/architecture addressing the problems described above.
- Implementing a lab test setup to verify the framework.
- Documenting the results in a thesis.
Contact:
David Kolden
david.kolden@ffi.no
90915641
Time synchronization of swarming agents
Swarming agents utilize information sharing (sensor measurement, localization data etc.) to coordinate their movements and actions. In more dynamic situations, e.g. tasks involving high velocities or sudden changes in the surroundings, the information shared between agents relies heavily on precise time synchronization. This could be done by having a time synchronization server installed on each agent and synchronizing these servers to the Global Positioning System (GPS) time. For swarming agents operating in denied environments over time, the reliance on satellite technology is a vulnerability that could reduce the performance and efficiency of the swarm drastically.
FFI wants to look at the scenario where a global master clock like the Coordinated Universal Time (UTC) received from GPS is unavailable and the swarming agents need to distribute and coordinate a local master clock among themselves. There exist various protocols for time synchronization such as IEEE 1588-2008 Time Synchronization Protocol (PTP), IEEE 802.1AS Time Synchronization Protocol, GPS Time Synchronization Protocol, IRIG-B Time Synchronization Protocol and Time Distribution Protocol (from European Space Agency). Many of these protocols are typically designed around a cabled network and handled by hardware in layer 2 of the Open Systems Interconnection (OSI) model. In the scenario described above, the time synchronization must be done through wireless networks with varying degrees of connectivity and latency. In addition, the system is inherently decentralized, thus making it hard to maintain a single master clock over time.
Some main aspects of the assignment could be:
- A study of relevant literature exploring both the state of the art of existing systems and the possibility of improvements.
- A comparison of existing protocols/systems.
- A design of a framework/architecture addressing the problems described above.
- Implementing a lab test setup to verify the framework.
- A study of the effect of different qualities of time synchronization on applications relying on this, e.g. tracking, mapping and sensor fusion between agents.
- Documenting the results in a thesis.
Contact:
David Kolden
david.kolden@ffi.no
90915641