Projects for Master Thesis - Open for academic period 2021-2022. 

Multi-modal Localization And Mapping in Perceptually-degraded Environments
Supervisors: Kostas Alexis |Keywords: SLAM, GPS-denied, Visually-degraded
Available
This thesis aims to develop a new class of Simultaneous Localization And Mapping (SLAM) algorithms that exploit the fusion of diverse modalities to enable the autonomous navigation inside GPS-denied and perceptually-degraded environments. In particular, the goal of fusion of LiDAR, visible-light and thermal cameras, as well as inertial sensors is considered. We envision the development of a factor-graph based optimization problem for integrating the diverse modalities, alongside appropriately designed multi-modal data association front-ends. The designed method will be tested onboard a flying robot integrating a both a 3D LiDAR and visible light cameras synchronized with a high quality Inertial Measurement Unit. Field experiments will be conducted in underground environments presenting diverse cases of degradation including geometric self-similarity, darkness and broadly lack of visual features.
Relevant video: ​​https://youtu.be/Tv6FoQI8h_I
Reinforcement Learning for Autonomous Exploration of Cluttered Environments using Collision-resilient Aerial Robots
Supervisors: Kostas Alexis |Keywords: Reinforcement Learning, Autonomy
Available
This thesis aims to develop a reinforcement learning approach applicable to continuous state and action spaces and partially observed Markov decision processes, wherein the reward for the agent is contingent upon the agent’s efficient exploration of the environment. The method should be able to utilize a sliding window of LiDAR (and/or camera) observations in order to identify reference paths that a) maximize the anticipated exploration reward, while b) minimizing the likelihood of a collision and necessarily ensuring that if a collision is bound to happen then the kinetic energy at contact-time is such that the collision-tolerant robot can remain safe. To enable robust performance in very long-term and large-scale deployments we seek to develop a method that does not assume a consistent online reconstructed 3D map of the environment but rather aim to encode the ability to “develop map memory” through recurrent networks. 
Data-driven Model Predictive Control for Collision-tolerant Aerial Robots
Supervisors: Kostas Alexis |Keywords: Aerial Robots, Control, Learning, Navigation
Available
This thesis aims to develop a novel controller for collision-tolerant aerial robot capable of affording impacts with the environment up to a certain speed. The envisioned control policy should bring together the strengths of Model Predictive Control and modern deep neural networks towards delivering a behavior that ensures the safety of the robot by avoiding any collision with the environment or ensuring that an impact will only happen with limited kinetic energy. To acquire data for the map of the environment, a 3D LiDAR will be used by the aerial robotic system delivering a rich depth representation of the robot’s surroundings. The developed solution will be evaluated both in simulation but also onboard a collision-tolerant aerial robot. The vision of this work is to deliver a control policy that exploits collision tolerance and thus allows for more agile and high-speed navigation inside highly confined and complex environments.​
Relevant video: ​https://youtu.be/LcsuX1bwI_8
Fast Frontier Detection for High-speed Exploration using Aerial Robots
Supervisors: Kostas Alexis | Keywords: Space & Subterranean Robots, Mechatronics, Control
Available
This thesis aims to research new methods on fast and computationally efficient detection of the mapped space in order to facilitate high-speed autonomous exploration using agile aerial robots. In particular, the goal is to support Micro Aerial Vehicles in missions of autonomous exploration and mapping in environments for which no prior knowledge is available. Exploiting an online occupancy map representation of the environment the task is to first detect the frontiers of the explored space and then identify a dynamics-aware collision-free path towards that point in the robot’s configuration space. We envision update rates that are fast enough for a robot to be able to fly with speeds as high as 5m/s, which in turn implies that the new path must be computed in very limited time by relying solely on the onboard computational resources.
Relevant video: ​https://youtu.be/ZvUedi5mzN8
Assured Learning-based Safe Navigation within Degraded and Constrained Environments
Supervisors: Kostas Alexis | Keywords: Reinforcement Learning
Available
This thesis aims to develop the new methods and technological solutions required to enable learning-based verifiably safe navigation of aerial robotic systems operating inside sensor-degraded and constrained/cluttered environments. In particular, a novel navigation policy, derived through assured reinforcement learning techniques, is proposed that will guide an autonomous flying system across reference waypoints, while autonomously ensuring a) the avoidance of 3D obstacles inside dense and narrow environments, b) sampling of data from the subsets of the environment that present interest to the robot’s mission, and c) accounting for the effects of localization uncertainty to the problem of safe navigation. To offer guarantees for field deployment, the envisioned work aims to introduce reachability analysis inside the training process and a verification tool to certify the safety and performance of the learned policy. The results will be demonstrated in the context of challenging field experiments involving the autonomous navigation inside ballast water tanks. 
Optical Flow-based Reinforcement Learning for Collision-free High-speed Navigation of Aerial Robots
Supervisors: Kostas Alexis | Keywords: Reinforcement Learning
Available
This thesis aims to develop a reinforcement learning approach that exploits vision-based optical flow for aerial robots in order to ensure high-speed navigation and collision-avoidance for aerial robots. The method should utilize a sliding window of camera observations from which a temporal cue of optical flow maps can be derived. This information stream should then be exploited by a reinforcement learning-based agent trained to maximize a reward reflecting how the system approaches desired destinations and ensures avoiding all obstacles on its way. To enable robust performance and high-speed navigation of lightweight Micro Aerial Vehicles we seek to exploit small cameras and limited onboard computational processing thus we propose the use of the intermediate representation of optical scene flow as opposed to directly training on the visual data input stream. The method is to be developed with the combination of simulation and real-life data and it is aimed to be tested with the “RMF” collision-resilient aerial robot (depicted in the Figure on the right) developed in the framework of the DARPA Subterranean Challenge.
Relevant video: ​https://youtu.be/LcsuX1bwI_8 
Mars Regolith Digger: Robotic system to autonomously excavate on Mars 
Supervisors: Kostas Alexis | Keywords: Space & Subterranean Robots, Mechatronics, Control
Available
In this work we envision the conceptual design, detailed study and miniaturized prototype realization of a robotic system capable of autonomous excavation of Mars regolith ground. Such a robotic system is to be designed to investigate new means for the future of autonomous preparation of a human base on Mars. Regolith will be used to develop the needed structural materials and thus its robotized excavation is essential. In this work the main focus is on the mechatronic design and control of this Mars Regolith Digger concept, while further study on its perception solution is also possible if time allows.
Hopping-and-Roving Robots for Martian Lava Tube Exploration
Supervisors: Kostas Alexis | Keywords: Space & Subterranean Robots, Mechatronics, Control
Available
In this thesis, the aim is to develop a new class of roving and jumping subterranean robotic explorers tailored to the needs and challenges of Martian Lava Tubes. The goal is to offer simultaneously: a) endurance exceeding 2h at a single charge and a range exceeding 10km, b) traversability of challenging terrain, and c) a small size and weight (<10kg). The latter is particularly important as weight highly affects the cost of the mission, while the size must be small enough for skylight access and teamed operation eliminates the possibility of a single-point-of-failure. To achieve such challenging design specifications, we envision a robotic unit involving a hopping-and-roving robot. This robot will rely on large wheels with high-torque motors attached onto a simple leg mechanism to help in traversing challenging terrain, and implement a jumping mechanism. Control of the wheel spin and the leg configuration will also facilitate attitude control during the flight-phase of each jump. Exploiting the reduced effects of gravity of Mars, approximately 38% that of Earth, the system can benefit from its very small weight to seamlessly overcome challenging obstacles through long jumps – an approach even more relevant due to the large size of lava tubes on Mars. Within the framework of the thesis we seek to develop the conceptual design and a prototype realization of such a hopping-and-roving robot.
Master project/theses topics in Underwater Robotic Autonomy
Supervisors: Kostas Alexis (NTNU), Eleni Kelasidi (SINTEF) | Keywords: Underwater Robots
Available
Four master thesis have been proposed for Unmanned Underwater Vehicles (UUVs) use in Aquaculture domain in collaboration with SINTEF Ocean. The master projects target localization, mapping, motion planning and advanced control for UUVs operating in dynamically changing environments such as fish farms where the underwater vehicles need to operate in the presence of alive fish and deformable structures:
  • Visual and hydro-acoustic data to identify the interaction between unmanned underwater vehicles and fish for real-time control of UUVs
  • Stochastic mapping for robotic operations in the presence of fish and deformable structures
  • Multi-modal underwater localization and mapping for UUVs operating in fish farms
  • Collision-free motion planning for operations in dynamically changing environments such as fish farms
  • Details available following this link. 
High Wind Transition for a Class of Vertical Take-Off and Landing Unmanned Aerial Vehicles
Supervisors: Kostas Alexis | Keywords: VTOL/Fixed-wing UAV Control
Available
This thesis aims to develop an optimized flight control and motion planning framework to enable operation in highly windy environments for a special class of Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs) capable of convertible flight, namely from multi-rotor mode to fixed-wing mode.
Real-time Vision-Aided Inertial Navigation for Vertical Take-Off and Landing Unmanned Aerial Vehicles During Critical Flight Stages
Supervisors: Kostas Alexis | Keywords: VTOL/Fixed-wing UAV, Visual-inertial Odometry
Available
This thesis aims to develop a robust vision-aided inertial navigation framework for a special class of Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs) during critical flight stages. The vehicle is capable of convertible flight, namely from multi-rotor mode to fixed-wing mode, and is used to transport blood samples between hospitals.
Three Dimensional Online Wind Estimation Onboard a Class of Vertical Take-Off and Landing Unmanned Aerial Vehicles
Supervisors: Kostas Alexis | Keywords: VTOL/Fixed-wing UAV, Wind estimation, AoA, Sideslip
Available
This thesis aims to develop an online estimation framework and sensor suite for full three dimensional wind estimation onboard a special class of Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs) capable of convertible flight, namely from multi-rotor mode to fixed-wing mode.
High Energy Landing for a Class of Vertical Take-Off and Landing Unmanned Aerial Vehicles
Supervisors: Kostas Alexis | Keywords: VTOL/Fixed-wing UAV Control
Available
This thesis aims to develop an optimized flight control and motion planning framework to enable high energy landing for a special class of Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs) capable of convertible flight, namely from multi-rotor mode to fixed-wing mode.
Agile collision-free navigation and path planning for an autonomous surface vehicle
Supervisors: Kostas Alexis | Keywords: Maritime robotic autonomy
Available
Develop an agile guidance system for an unmanned surface vehicle (USV) platform that is capable of safe and efficient navigation to stochastically distributed survey areas in the presence of dynamic entities (moving vessels) and stationary entities (island, large debris).  Collision avoidance must follow the Convention on the International Regulations for Preventing Collisions at Sea (COLREGs) adapted for autonomous vessels.
Real-Time Strategy game-like Control of Very Large Robotic Teams
Supervisors: Kostas Alexis | Keywords: Path planning, simulation, AI
Available
This thesis deals with the goal of developing the software framework necessary that would allow the commanding of very large fleets [100+] of robotic systems in a manner analogous to that experienced when playing real-time strategy games (RTS). In such video games, the user has access to immediate control of diverse “units” which can be commanded to seamlessly execute a collection of missions including exploration, surveillance and patrolling. The effective result is that of a “mosaic of agents” within which every unit contributes to the final goal but none is absolutely necessary. Thus resilience in the ability to execute the the user’s command is achieved by the aggregation of multiple units and the simplicity to operate them. Motivated by this we seek to develop on top of successful simulators in robotics in order to deliver analogous functionality. As modern robotic simulators provide a level of abstraction that allows code developed in them to be API-wise compatible with real robotic systems, this work is a significant prior step towards seamless control of very large fleets of real-life robotic systems in the context of challenging applications. In terms of types of robots, ground and aerial robots such as the legged system and the collision-tolerant Micro Aerial Vehicle presented above are of immediate interest.​
Online extrinsic recalibration of sensors on a serial-link robot
Supervisors: Kostas Alexis (NTNU), Esten Ingar Grøtli (SINTEF) | Keywords: SLAM
Available
Accurate localization is a prerequisite for robots in and indoor or GNSS-denied environment. When using multiple sensors, information from each sensor must be fused to provide the best possible estimate. However, to fuse the measurements, the sensors need to be accurately calibrated. This is a costly and time-consuming process, and often sensors need to be re-calibrated. The aim of this project is to re-calibrate sensors on-the-fly using perceived features from the environment. A first step of this project will consist in placing multiple cameras (see image), LiDAR and IMU on a robot and ensure that the sensors are time synchronized and calibrated. The second step is to adapt an available SLAM pipeline to this snake robot. The third step is the development of an online solution for recalibration, and testing of this against traditional solutions for calibration.​
Through-the-ground Communications for Subterranean Robots
Supervisors: Kostas Alexis | Keywords: Subterranean Robots, Communications
Available
This thesis aims to design and prototype a solution for through-the-ground communications for robotic systems deployed to explore subterranean environments such as cave networks and underground mines. Such underground environments can be several meters to hundreds of meters below ground level thus rendering communications to an operating autonomous from extremely challenging to impossible. Traditional through-the-ground communications are typically bulky and heavy as they rely on extremely long antennas due to the necessary long wavelengths involved. At the same time such communications come with additional limitations (e.g., in terms of the possible bandwidth). In this work we seek to investigate the design space that would allow to realize a reasonably lightweight communications solution enabling through-the-ground communications with sufficient bandwidth to exchange minimalistic messages relating to a) the current robot pose, b) possible detections of objects of interest, and c) issuing of emergency commands to stop the robot or request it to return to home. Prototype realization of the envisioned solution is an essential component of this work, including the modules onboard robotic systems and the module on the ground, alongside their possibly different antennas.
Soft Deep Sea Robotic Explorer
Supervisors: Kostas Alexis | Keywords: Underwater robots, Soft robotics, Control
Available
In this thesis, the goal is to develop the design and prototype realization of a soft robotic system capable of operating in deep sea waters to perform scientific exploration and sampling missions. The design aims to address two challenges, namely a) what material selection and/or fabrication can allow the system to sustain the pressure at significant depths, and b) the actuation mechanism for throttling and steering the robot. As soft materials are to be used, this opens the way for small size robots that exploit such mechanical properties in order to sustain the high pressure without resorting to heavy and rather bulky designs. To draw inspiration for the design you are advised to look at species that naturally live in such deep sea environments. The research will naturally need to combine skills and interest in mechatronic design, understanding of material properties, modeling and control.​
Image from sciencenews.org.
Paper: 
Li, G., Chen, X., Zhou, F., Liang, Y., Xiao, Y., Cao, X., Zhang, Z., Zhang, M., Wu, B., Yin, S. and Xu, Y., 2021. Self-powered soft robot in the Mariana Trench. Nature, 591(7848), pp.66-71.
(not work of our lab)

Previous Academic Period 

Flight Control and Motion Planning for a Class of VTOL UAVs
Supervisors: Kostas Alexis
Assigned
This thesis aims to develop optimized flight control and motion planning algorithms to enable the  robust and high performance flight of a class of Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs) capable of convertible flight, namely from rotorcraft mode to fixed-wing mode. The research will emphasize on robust flight control especially against wind disturbances, alongside the generation of optimized trajectories that account for estimates of the wind field across the direction of motion. The method will be implemented on the PX4 open-source autopilot software and deployed onboard a prototype VTOL UAV. Details here. 
Resilient Flight Control for a Collision-tolerant Aerial Robot
Supervisors: Kostas Alexis
Assigned
This thesis aims to develop a novel control strategy tailored to a Resilient Micro Flyer implementing a collision-tolerant frame. The method should account and exploit the collision tolerance of not only to passively mitigate the risks of a collision but also to a) actively modify the control commands when a collision-event is detected, and b) to intentionally decide if the robot can navigate an environment better by maintaining smooth contact with a surrounding surface. Alongside the controller design, the implementation of a collision detection algorithm relying on the onboard Inertial Measurement Unit data is also necessary. 
Blazing Fast Exploration Path Planning for Subterranean Flying Robots
Supervisors: Kostas Alexis
Assigned
​This project aims to develop a path planning strategy for autonomous aerial robots in order to explore large-scale underground environments, such as mines and cave networks, at rapid speeds. Building on top of a set of previously developed planning methods of our team, the goal is to address two core limitations relating to the environment representation and the information gain formulation used to derive optimized exploration paths and thus allow to efficiently calculate dynamic and agile trajectories allowing high-speed exploration. The latter is a necessary condition if battery- and broadly resource-constrained robots are to be able to efficiently explore the often km-long underground environments. Check examples of our previous work. 
Collision Resilient Navigation for Aerial Robots in Confined Environments
Supervisors: Kostas Alexis, Christos Papachristos
Assigned
​This research aims to enhance the navigation abilities of aerial robots in confined/narrow environments by enabling them to stably establish contact with surfaces of their environment. More specifically, a Micro Aerial Vehicle will be enhanced with specialized mechanisms for physical interaction (extensions) and a software framework for control in confined spaces by exploiting contact. Force feedback at the end effectors will facilitate stable and sustainable physical interaction. This research direction plans to radically change how flying robots navigate through narrow environments such as ore-passes or manholes.
Reactive Collision Avoidance for Aerial Robots Navigating in Underground Environments
Supervisors: Kostas Alexis
Assigned
This research aims to investigate a last-resort collision-avoidance mechanism that is implemented onboard an aerial robot aiming to navigate complex underground settings. The method should employ different sensing modalities in order to provide robustness and satisfactory performance even in visually-degraded conditions. In particular, the combination of visible camera data and LiDAR ranging is considered as a starting point. Ideally, the designed solution should not assume that a reliable pose and map estimate is available and enable reactive avoidance even in the most degenerate cases.
On the role of Levy Flights towards Efficient Memoryless Robotic Search
Supervisors: Kostas Alexis
Available
Autonomous robots are commonly tasked with the problem of area exploration and search for certain targets or artifacts of interest to be tracked. Traditionally, the problem formulation considered is that of complete search and thus - ideally - identification of all targets of interest. An important problem however which is not often addressed is that of time-efficient memoryless search under sparse rewards that may be worth visited any number of items. In this work we want to address the largely understudied problem of optimizing the "time-of-arrival" or "time-of-detection" to robotically search for sparsely distributed rewards (detect targets of interest) within large-scale environments and subject to memoryless exploration. 

​Relevant "Working" Paper: Download
Simulation Environment for Subterranean Robotics
Supervisors: Kostas Alexis, Christos Papachristos
Available
This project aims to develop a comprehensive simulation environment for testing ground and flying robots for autonomous navigation in subterranean environments. It will be based on Gazebo and ideally should also allow for some type of Hardware-In-the-Loop functionality. It will develop on top of the Gazebo-based simulator of subterranean environments provided by DARPA. 
Understanding the Vision System of Underground Species
Supervisors: Kostas Alexis
Available
In this project we question what is special and specific to the vision system of specials (from animals to insects) living underground. Good examples include the wolf spider or beavers. The emphasis is on literature study on the specific domain and derivation of conclusions on how certain principles may apply to robotic vision systems both in the sense of hardware and algorithms. It corresponds to a project that will lay the ground for many subsequent investigations to follow. 

Own ideas - high risk projects!

Do you have your own idea about a robotics project? Are you willing to discuss a high-risk project with the understanding that things might not always work? Contact me and schedule a meeting!