Student Projects

DISAL offers a range of student projects its three main areas of expertise: Distributed Robotic Systems, Sensor and Actuator Networks, and Intelligent Vehicles. For more information about supervision guidelines and different types of student projects please refer to this page.

Spring Semester 2022-2023

Automatic Design of Behaviors for Khepera IV Robots Emulating Automotive Vehicles

 

Assigned to David Fontes Junqueira

With the increased deployment of mobile robotics in a multitude of real-life scenarios, robots must face more and more complex situations. However, every robotic platform needs to be carefully programmed to fulfill its task, which, today, is often still done manually. Such programming operation usually represents an intense time investment and requires high expertise to eventually achieve the targeted performances in a real deployment, even more so if the application involves a multi-robot system. There are multiple approaches for the automatic generation of controllers (or specific components), whereas the nature of the control components ranges from Probabilistic Finite State Machines (PFSM) [1] to Artificial Neural Networks (ANNs) [2].  However, powerful machine-learning techniques are necessary to automatically generate any of these controllers.

This project aims to automatically generate basic behaviors which can then be leveraged using behavioral arbitrators for more complex scenarios. Starting from the result of a previous semester project and the recent work in [3], this project aims to extend the generated basic behaviors to multi-robot situations using Webots, a high-fidelity, open-source robotics simulator. In particular, a library of basic behaviors will be generated for Khepera IV robots. The combined effectiveness of the resulting behavior-based control architecture will be demonstrated in a realistically simulated traffic scenario, with the Khepera IV robots emulating automotive vehicles. The expected effort and outcome of this work will be adjusted based on the type of project.

Recommended type of project: Semester project / Master project

Work breakdown: 30% theory, 40% software, 30% experimentation           

Prerequisites: broad interest in machine learning, neural networks, control algorithms and robotics, good command of C/C++

Keywords: machine learning, robot control, automatic design and optimization

Contact: Cyrill Baumann

Reference:
[1] C. Baumann and A. Martinoli, “A Noise-Resistant Mixed-Discrete Particle Swarm Optimization Algorithm for the Automatic Design of Robotic Controllers” IEEE Congress on Evolutionary Computation, 2022
[2] Duarte M., Sancho Moura O., and Anders Lyhne C., “Evolution of Hybrid Robotic Controllers for Complex Tasks” Journal of Intelligent & Robotic Systems, 2015, 78: 463
[3] Archit Sharma, Michael Ahn, Sergey Levine, Vikash Kumar, Karol Hausman, Shixiang Gu, “Emergent Real-World Robotic Skills via Unsupervised Off-Policy Reinforcement Learning”,  Robotics: Science and Systems, 2020

Discrete Optimization for Automatic Design of Behaviors for Khepera IV Robots

Assigned to Justin Manson

With the increased deployment of mobile robotics in a multitude of real-life scenarios, robots must face more and more complex situations. However, every robotic platform needs to be carefully programmed to fulfill its task, which, today, is often still done manually. Such programming operation usually represents an intense time investment and requires high expertise to eventually achieve the targeted performances in the real deployment, even more so if the application involves a multi-robot system. There are multiple approaches for the automatic generation of controllers (or specific components), whereas the nature of the control components ranges from Probabilistic Finite State Machines (PFSM) [1] to Artificial Neural Networks (ANNs). However, powerful machine-learning techniques are necessary to automatically generate any of these controllers.

This project aims to improve a recently proposed separation of a PFSM synthesis approach into a discrete (categorical) and continuous optimization problem [2], by exploring alternative algorithms for the discrete optimization. Starting from a literature search on the state-of-the-art of discrete (categorical) optimization algorithms, this project then aims to implement the most promising solution(s). After an initial comparison using benchmark functions, the most promising algorithm(s) will then be compared to the current solution consisting of a Mixed-Discrete Particle Swarm Optimization [3] algorithm on a concrete PFSM synthesis problem for Khepera IV robots. The expected effort and outcome of this work will be adjusted based on the type of project.

Recommended type of project: Semester project / Master project

Work breakdown: 20% theory, 40% software, 40% experimentation           

Prerequisites: broad interest in machine learning, optimization algorithms, good command of python or C/C++

Keywords: machine learning, optimization, automatic design and optimization

Contact: Cyrill Baumann

Reference:
[1] C. Baumann and A. Martinoli, “A Noise-Resistant Mixed-Discrete Particle Swarm Optimization Algorithm for the Automatic Design of Robotic Controllers”, IEEE Congress on Evolutionary Computation, 2022, DOI: 10.1109/CEC55065.2022.9870229 (9 pages).
[2] C. Baumann, H. Birch, A. Martinoli, “Leveraging Multi-Level Modelling to Automatically Design Behavioral Arbitrators in Robotic Controllers”, IEEE/RSJ International Conference on Intelligent Robots and Systems, 2022, pp. 9318-9325.
[3] Chowdhury, S., Tong, W., Messac, A. et al. A mixed-discrete Particle Swarm Optimization algorithm with explicit diversity-preservation. Structural and Multidisciplinary Optimization, 2013, pp. 367-388.

Heterogeneous Multi-Robot Gas Distribution Mapping and Source Localization

Gas Distribution Mapping (GDM) and Gas Source Localization (GSL) are useful techniques for real world applications such as environmental monitoring. Implementations so far have focused on 2D mapping and localization [1][2], but a 3D approach would be very valuable, given the tridimensional nature of the plume. Researchers at DISAL are focusing on exploiting quadrotors for gas detection missions in 3D [4] [5]. Moreover, using an array of robotic assets with different properties (e.g., UAVs, wheeled robots), could be beneficial for improving the final gas map.

This project will explore the benefits of a heterogeneous multi-robot architecture in a mission that aims at achieving simultaneous gas distribution mapping and gas source localization. An appropriate heterogeneous multi-robot architecture will be selected, and algorithms will be developed to achieve both tasks at the same time. The student will focus on cooperative and collaborative strategies to enhance the performance of the system and on developing strategies to best exploit the characteristics of each robotic platform.

 

Recommended type of project: semester/master project

Work breakdown: 40% theory, 40% simulation, 20% experiments

Prerequisites: Broad interest in robotics, very good programming skill (Python/C/C++), knowledge of ROS, strong interest in research

Keywords: gas source mapping, quadrotors, multi-robot system

Contact: Chiara Ercolani

References:

[1] Lilienthal, Achim, and Tom Duckett. “Creating gas concentration gridmaps with a mobile robot.” Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems Vol. 1. 2003, p. 118-123.

[2] Lilienthal, Achim J., et al. “A statistical approach to gas distribution modelling with mobile robots-the kernel dm+ v algorithm.” Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 2009, p. 570-576.

 [3] Ercolani, Chiara and Martinoli, Alcherio “3D Odor Source Localization using a Micro Aerial Vehicle: System Design and Performance Evaluation” Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, 2020, p. 6194-6200

[4] Ercolani, Chiara, et. al. “Clustering and Informative Path Planning for 3D Gas Distribution Mapping: Algorithms and Performance Evaluation” Robotics and Automation Letters, 2022, p. 5310-5317.

Particle Filters for Gas Source Localization in Cluttered Environments

         

Deploying robots for Gas Source Localization (GSL) [1] tasks under hazardous scenarios can significantly reduce the risk for human and animals typically engaged in such missions. The stochastic nature of gas dispersion makes GSL a challenging task. Gas sensing using mobile robots focuses primarily on simplified scenarios, with a current trend towards tackling more complex environments. The presence of obstacles in the environment influences the process of gas dispersion and increases spatial stochasticity in the gas measurements. The GSL problem can be formulated as a state estimation problem, and particle filter algorithms are powerful tools to solve state estimation [4] tasks from scattered and noisy measurements.

The student will explore an adaption of a particle filter algorithm [2] for GSL by investigating the correlation between the gas source parameters (including location) and gas measurements in a cluttered environment. Furthermore, for the robot navigation, the influence of the environmental structure [3] on the measurements needs to be considered for path planning. The performance of the algorithm will have to be thoroughly evaluated in a high-fidelity robotic simulator (Webots).

Recommended type of project: semester project, master project

Work breakdown: 60% theory, 40% simulation

Prerequisites: passion on solving issues, broad interest in robotics, good programming skill (Python/C++), experience with Webots will be an asset

Keywords: gas source localization, navigation, state estimation, path planning, Webots

Contact: Wanting Jin

References:

[1] Francis, A., Li, S., Griffiths, C. & Sienz, J. (2022) Gas source localisation and mapping with mobile robots: A review. Journal of Field Robotics, 1– 33.

[2] J. R. Bourne, E. R. Pardyjak and K. K. Leang, “Coordinated Bayesian-Based Bioinspired Plume Source Term Estimation and Source Seeking for Mobile Robots,” in IEEE Transactions on Robotics, vol. 35, no. 4, pp. 967-986, Aug. 2019.

[3] Rhodes, C., Liu, C., Westoby, P. and Chen, W.H., 2022. Autonomous search of an airborne release in urban environments using informed tree planning. Autonomous Robots, pp.1-18.

[4] F. Rahbar, A. Marjovi and A. Martinoli, “An Algorithm for Odor Source Localization based on Source Term Estimation,” 2019 International Conference on Robotics and Automation (ICRA), 2019, pp. 973-979, doi: 10.1109/ICRA.2019.8793784.

Sensor Network Development for Gas Source Localization

Assigned to: Michael Wayne Freeman

In environmental monitoring tasks, static sensor networks [1][2] are highly complementary assets compared to mobile platforms, given their higher energetic autonomy and measurement quality. In addition, they benefit from the ability to make synchronous parallel measurements in space, enabling to capture time-variant features of the environment. A new static sensor node for gas source localization tasks has been designed at DISAL. It consists of four modules: a core computational board (M5Stack Core2 [3]), a gas sensing module, a wind sensing module, and a battery pack. To deploy an extendable and manageable sensor network that comprises a sufficient number of sensor nodes in real applications, some additional factors must be considered: first, the energetic autonomy needs to be sufficient for long-run experiments; second, the communication protocol needs to be reliable and extendable.

To solve these problems, a system that controls the wake-up and sleep modes of all sensors and manages the measurements through wireless communication is required. As communication protocol, we have chosen MQTT [4], a lightweight publish/subscribe messaging system designed for setting up bidirectional communication between the host machine and multiple devices with minimal network bandwidth. In order to facilitate a hybrid-mobility sensing system [5] that coordinates the static sensor network with robotic sensing assets, a management and communication system for the sensor network will be designed under the ROS framework. The student will work with real hardware and test system performance in physical gas source localization scenarios.

Recommended type of project: semester project, master project

Work breakdown: 30% theory, 70% development

Prerequisites: passion on solving issues, broad interest in sensor networks, good programming skill (Python/C/ Arduino), knowledge of ROS will be an asset

Keywords: Sensor Network, Arduino, ROS, MQTT

Contact: Wanting Jin

References:

[1] J. Burgués, V. Hernández, A. J. Lilienthal, and S. Marco, “Gas distribution mapping and source localization using a 3D grid of metal oxide semiconductor sensors,” Sensors and Actuators B: Chemical, vol. 304, p. 127309, Feb. 2020.

[2] A. Marjovi, A. Arfire and A. Martinoli, “High Resolution Air Pollution Maps in Urban Environments Using Mobile Sensor Networks,” 2015 International Conference on Distributed Computing in Sensor Systems, 2015, pp. 11-20, doi: 10.1109/DCOSS.2015.32.

[3] https://docs.m5stack.com/

[4] https://mqtt.org/

[5]  W. C. Evans, D. Dias, S. Roelofsen, and A. Martinoli, “Environmental field estimation with hybrid-mobility sensor networks,” in 2016 IEEE International Conference on Robotics and Automation (ICRA). Stockholm: IEEE, May 2016, pp. 5301–5308.

State estimation and localization for autonomous UAVs

Autonomous Micro Aerial Vehicles (MAVs) require a complete state estimate in order to stabilize themselves, and navigate their environment. Commonly, MAVs are equipped with inertial sensors (IMU) and GNSS receivers, whose information is fused to provide a complete state estimate [1]. However, GNSS signals can be partially or completely unavailable. For this reason, visual localization solutions have been developed, most notably Visual Inertial Odometry (VIO) [2], that solely rely on IMU and image data and can be run onboard in real-time. However, VIO is usually a dead-reckoning solution and is subject to drift. Simultaneous Localization and Mapping (SLAM) [3] allows to mitigate this issue but can be computationally (very) expensive. It is therefore usually not possible to leverage this kind of algorithm directly onboard a MAV, given its limited resources.

This project therefore aims at evaluating how visual SLAM can be used for the absolute long-term localization of an autonomous MAV performing onboard state estimation, using VIO for direct stabilization, or directly using a SLAM solution without VIO. The software configuration (for instance VIO onboard, SLAM run on ground station, or only SLAM onboard, etc…) shall be investigated, as well as the fusion of both sources of estimation if two pipelines are run concurrently. To this end, a literature review shall be carried out to gain better insight on the available open-source visual SLAM implementations, and suitable candidates should be selected and tested in simulation, such as ORB-SLAM3 [4]. Faithful drone simulations will be performed using the Webots simulator [5]. If time permits, tests with real drones can be performed.

Recommended type of project: semester project

Work breakdown: 40% theory, 60% simulation

Prerequisites: Broad interest in robotics, very good programming skills (C/C++, Python). Knowledge in ROS is required, but only recommended for Webots and git. Having attended the course “Aerial robotics” or similar is recommended.

Keywords: Mobile robotics, state estimation, localization, MAV, simulation, Webots, quadrotors, computer vision, ROS.

Contact: Lucas Wälti

References:

[1] Mahony, Robert, Vijay Kumar, and Peter Corke. “Multirotor Aerial Vehicles: Modeling, Estimation, and Control of Quadrotor.” IEEE Robotics & Automation Magazine 19, no. 3 (September 2012): 20–32. https://doi.org/10.1109/MRA.2012.2206474.
[2] Scaramuzza, Davide, and Zichao Zhang. “Visual-Inertial Odometry of Aerial Robots.” ArXiv:1906.03289 [Cs], June 14, 2019. http://arxiv.org/abs/1906.03289.
[3] Grisetti, G, R Kummerle, C Stachniss, and W Burgard. “A Tutorial on Graph-Based SLAM.” IEEE Intelligent Transportation Systems Magazine 2, no. 4 (2010): 31–43. https://doi.org/10.1109/MITS.2010.939925.
[4] Campos, Carlos, Richard Elvira, Juan J. Gómez Rodríguez, José M. M. Montiel, and Juan D. Tardós. “ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM.” IEEE Transactions on Robotics 37, no. 6 (December 2021): 1874–90. https://doi.org/10.1109/TRO.2021.3075644.
[5] “Webots: Robot Simulator.” Accessed October 6, 2022. https://cyberbotics.com/.

Point cloud segmentation of infrastructural steel elements

     

Community-critical steel infrastructure, such as energy transmission and telecommunication lines, steel bridges, offshore rigs, etc… require regular inspection, in order to guarantee the integrity of the structure and good operations. Current inspection approaches usually require the use of heavy equipment, as well as visual inspection performed by technicians, which exposes them to potentially dangerous situations. Furthermore, inspections usually imply down-time for the structure under inspection, which is costly and inefficient. The main interest is therefore to identify fatigue signs within a structure, without requiring to stop its usage. Drones endowed with visual sensing capabilities are potentially good candidates to achieve this goal.

This project aims at understanding and modeling online the geometry of steel infrastructures from 3D measurements (typically generated by a stereo camera pair or Time of Flight camera) in real time. The detection can possibly be coupled with computer vision techniques. A suitable representation should be selected (raw point cloud, voxel grid [2], geometry primitives fitting, …). Ways of segmenting the point cloud shall also be explored [3], by possibly leveraging the characteristic of the steel structure. The purpose for the perception of the geometry is to enable autonomous navigation along and within a steel infrastructure. Existing tools such as the Point Cloud Library (PCL [4]) shall be used. It will likely be required to work with ROS, for which previous experience is preferable. Everything will be performed in simulation using the Webots simulator [1]. If time permits, tests with real drones can be performed.

Recommended type of project: semester project

Work breakdown: 50% theory, 50% simulation

Prerequisites: Broad interest in robotics, very good programming skills (C/C++, Python). Knowledge in control theory, ROS, Webots, git. Having attended the course “Aerial robotics” or similar is recommended.

Keywords: Webots, simulation, quadrotors, auto-pilot, flight controller, point cloud, 3D, segmentation.

Contact: Lucas Wälti

References:

[1] Webots: https://cyberbotics.com/doc/reference/index

[2] Oleynikova, Helen, Zachary Taylor, Marius Fehr, Roland Siegwart, and Juan Nieto. “Voxblox: Incremental 3D Euclidean Signed Distance Fields for on-Board MAV Planning.” In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 1366-73, 2017. https://doi.org/10.1109/IROS.2017.8202315.

[3] Nguyen, Anh, and Bac Le. “3D Point Cloud Segmentation: A Survey.” In 2013 6th IEEE Conference on Robotics, Automation and Mechatronics (RAM), 225-30, 2013. https://doi.org/10.1109/RAM.2013.6758588.

[4] Point Cloud Library (PCL): https://pointclouds.org/ 

Fall Semester 2023-2024

The student projects for the next spring semester will be posted at the beginning of April 2023.