Periodic Reporting for period 2 - AirSens (High-Accuracy Indoor Tracking and Augmented Sensing using Swarms of UAVs)
Reporting period: 2019-08-01 to 2020-07-31
In the future, the low-aerial space will be crowded by unmanned aerial vehicles (UAVs) as the tendency is to make them increasingly smaller, smarter, and cheaper. Such peculiarities can open an unimaginable set of new applications thanks to the high flexibility and reconfigurability of UAVs. Nonetheless, the idea of having swarms of UAVs in future cities might be accepted with difficulty by the citizens because of their potential malicious use.
In this context, the overall goal of "AirSens" is to develop advanced radio signal processing techniques to be applied for high-accuracy tracking and sensing operations performed by UAVs, acting as distributed wireless sensor networks in mixed indoor/outdoor environments.
An example of considered scenario is an emergency situation in which UAVs constitute a temporary positioning infrastructure for mobile nodes located inside a building (e.g. firefighters). Another application of interest is the tracking of anomalous UAVs in critical safety areas. Going a step forward, one could easily imagine a setting where the UAVs enter a building with harsh propagation conditions, with many obstacles and a drastically reduced space for maneuvering. In this situation, UAVs should self-localize before being able to track users and to map the surrounding environment (infrastructure-less localization and mapping).
In all cases, the formation-navigation control of the UAV-swarm is based on the optimization of an information-theoretic cost function for seeking informative measurements and, hence, better estimating the parameters of interest.
To sum up, the main scientific objectives of "AirSens" are: (i) Analysis of the tracking performance using UAV-swarms; (ii) Assessment of the mapping performance of UAV-swarms; (iii) Control design and swarm intelligence algorithms.
By pursuing these objectives, AirSens deals with topics related to some of the societal challenges outlined by Horizon 2020, as for example:
(i) "Safety & Security". The swarm can be helpful for user guidance in unknown and dangerous environments. For example, in scarce visibility conditions, the swarm can guide rescuers in navigating buildings, or in rescuing human beings;
(ii) "Secure societies". The UAV swarm can be intended as “UAVs patrolling UAVs” and, thus, it can be useful for detecting and tracking malicious UAVs, especially in areas with potential safety issues, or it can be used in surveillance applications.
In this context, the overall goal of "AirSens" is to develop advanced radio signal processing techniques to be applied for high-accuracy tracking and sensing operations performed by UAVs, acting as distributed wireless sensor networks in mixed indoor/outdoor environments.
An example of considered scenario is an emergency situation in which UAVs constitute a temporary positioning infrastructure for mobile nodes located inside a building (e.g. firefighters). Another application of interest is the tracking of anomalous UAVs in critical safety areas. Going a step forward, one could easily imagine a setting where the UAVs enter a building with harsh propagation conditions, with many obstacles and a drastically reduced space for maneuvering. In this situation, UAVs should self-localize before being able to track users and to map the surrounding environment (infrastructure-less localization and mapping).
In all cases, the formation-navigation control of the UAV-swarm is based on the optimization of an information-theoretic cost function for seeking informative measurements and, hence, better estimating the parameters of interest.
To sum up, the main scientific objectives of "AirSens" are: (i) Analysis of the tracking performance using UAV-swarms; (ii) Assessment of the mapping performance of UAV-swarms; (iii) Control design and swarm intelligence algorithms.
By pursuing these objectives, AirSens deals with topics related to some of the societal challenges outlined by Horizon 2020, as for example:
(i) "Safety & Security". The swarm can be helpful for user guidance in unknown and dangerous environments. For example, in scarce visibility conditions, the swarm can guide rescuers in navigating buildings, or in rescuing human beings;
(ii) "Secure societies". The UAV swarm can be intended as “UAVs patrolling UAVs” and, thus, it can be useful for detecting and tracking malicious UAVs, especially in areas with potential safety issues, or it can be used in surveillance applications.
The main activities conducted for attaining the scientific objectives of “AirSens” are:
1. Tracking using UAVs: The first step was the derivation of a suitable figure of merit to assess the tracking performance of a UAV network, and to be used as a cost function for the navigation task. To this end, the localization fundamental limits were derived for a scenario where UAVs form a flexible network of reference nodes and exchange heterogeneous measurements. The second step was to develop different Bayesian target tracking algorithms (e.g. Kalman and particle filters), compliant with UAV mobility and with an observation model that encloses the information sensed by the swarm. Near-field signal and observation models were also considered to guarantee good performance in absence of synchronization.
2. Mapping using UAVs: An investigation of indoor mapping capabilities was conducted using a single mobile radar moving in indoors, able to collect measurement via beam-steering. Bayesian algorithms were compared considering real-world and simulated data. Mapping coverage and accuracy were maximized using reinforcement learning.
3. UAV Control: A decentralized control law was designed to estimate the next location of each UAV to best estimate the parameters of interest. Two different metrics were considered: a metric assessing the performance of a specific estimator; and a metric based on the fundamental limits (Fisher Information analysis). Secondly, a reinforcement learning approach was used where UAVs learn the optimal navigation policy by interacting with the environment.
As regards the training objectives of the project, the following steps have been performed:
1. Research: The researcher had the opportunity to collaborate with her project supervisors, who are top-class professors in the field of localization and signal processing. Together with them, she organized face-to-face meetings and Skype calls to report the project activities.
2. Networking: She has expanded her network of collaborations thanks to meeting organized with colleagues working in the same research fields (e.g. during regular meetings or for invited talks). In the context of an industrial collaboration, she assisted to a real-world demonstration of UAV navigation in indoors.
3. Talks: The researcher has attended several seminars held by top-class professors in her same research domain, and webinars on project proposal preparation held by the National Contact Point. She presented her research at three international conferences and at two panel meetings on artificial intelligence held at the European Research Council Executive Agency (REA).
4. Results: She has also focused on preparing and submitting contributions to scientific peer-reviewed journal papers and conferences proceedings. During the project, she published 5 journal papers, 5 conference proceedings, and 4 papers are currently under review. She is currently guest editor of Special Issues “Latest Advances on UAV Networks” and “Artificial Intelligence for Wireless Communications” for Open Access Journals by MDPI.
1. Tracking using UAVs: The first step was the derivation of a suitable figure of merit to assess the tracking performance of a UAV network, and to be used as a cost function for the navigation task. To this end, the localization fundamental limits were derived for a scenario where UAVs form a flexible network of reference nodes and exchange heterogeneous measurements. The second step was to develop different Bayesian target tracking algorithms (e.g. Kalman and particle filters), compliant with UAV mobility and with an observation model that encloses the information sensed by the swarm. Near-field signal and observation models were also considered to guarantee good performance in absence of synchronization.
2. Mapping using UAVs: An investigation of indoor mapping capabilities was conducted using a single mobile radar moving in indoors, able to collect measurement via beam-steering. Bayesian algorithms were compared considering real-world and simulated data. Mapping coverage and accuracy were maximized using reinforcement learning.
3. UAV Control: A decentralized control law was designed to estimate the next location of each UAV to best estimate the parameters of interest. Two different metrics were considered: a metric assessing the performance of a specific estimator; and a metric based on the fundamental limits (Fisher Information analysis). Secondly, a reinforcement learning approach was used where UAVs learn the optimal navigation policy by interacting with the environment.
As regards the training objectives of the project, the following steps have been performed:
1. Research: The researcher had the opportunity to collaborate with her project supervisors, who are top-class professors in the field of localization and signal processing. Together with them, she organized face-to-face meetings and Skype calls to report the project activities.
2. Networking: She has expanded her network of collaborations thanks to meeting organized with colleagues working in the same research fields (e.g. during regular meetings or for invited talks). In the context of an industrial collaboration, she assisted to a real-world demonstration of UAV navigation in indoors.
3. Talks: The researcher has attended several seminars held by top-class professors in her same research domain, and webinars on project proposal preparation held by the National Contact Point. She presented her research at three international conferences and at two panel meetings on artificial intelligence held at the European Research Council Executive Agency (REA).
4. Results: She has also focused on preparing and submitting contributions to scientific peer-reviewed journal papers and conferences proceedings. During the project, she published 5 journal papers, 5 conference proceedings, and 4 papers are currently under review. She is currently guest editor of Special Issues “Latest Advances on UAV Networks” and “Artificial Intelligence for Wireless Communications” for Open Access Journals by MDPI.
The proposed dual-functional UAV network for joint tracking and navigation is novel per se in the scientific panorama and, hence, no studies are available in the literature. In the following, we summarize the progress made beyond the state of the art in the context of the project.
- "UAVs as a Dynamic Sensor Network (DSN) for Tracking and Sensing": Wireless sensor networks for target tracking or environment monitoring have been investigated mainly in the context of fixed infrastructures. For example, to improve the safety of critical areas in relation to the presence of malicious drones, one can deploy ad-hoc terrestrial and fixed radar systems. Unfortunately, these systems fail in harsh environments due to obstacles that prevent the reception of the signal by terrestrial radars.
In this context, one of the breakthroughs of the project is to propose the design of a DSN, where sensors are flying on-board UAVs and, hence, provide a privileged point-of-view for collecting informative measurements about the environment. In order to realize the UAV-DSN, a promising solution is represented by mm-wave frequency-modulated continuous-wave radars because of the possibility to miniaturize the entire system. The same radar used for target tracking can be also exploited for mapping purposes.
- "UAV swarm control": The formation-navigation goal of the DSN is that of maximizing an information measure by seeking more informative measurements for better estimating the state of the system (information-seeking). In this sense, many research contributions have focused on optimal sensor placement and control, but they neglect the latency of the network, which becomes critical especially when the control is decentralized. In contrast to common approaches, one of the achievements of this project is to develop a control for UAVs able to assess a trade-off between localization accuracy and communication constraints.
- "UAVs as a Dynamic Sensor Network (DSN) for Tracking and Sensing": Wireless sensor networks for target tracking or environment monitoring have been investigated mainly in the context of fixed infrastructures. For example, to improve the safety of critical areas in relation to the presence of malicious drones, one can deploy ad-hoc terrestrial and fixed radar systems. Unfortunately, these systems fail in harsh environments due to obstacles that prevent the reception of the signal by terrestrial radars.
In this context, one of the breakthroughs of the project is to propose the design of a DSN, where sensors are flying on-board UAVs and, hence, provide a privileged point-of-view for collecting informative measurements about the environment. In order to realize the UAV-DSN, a promising solution is represented by mm-wave frequency-modulated continuous-wave radars because of the possibility to miniaturize the entire system. The same radar used for target tracking can be also exploited for mapping purposes.
- "UAV swarm control": The formation-navigation goal of the DSN is that of maximizing an information measure by seeking more informative measurements for better estimating the state of the system (information-seeking). In this sense, many research contributions have focused on optimal sensor placement and control, but they neglect the latency of the network, which becomes critical especially when the control is decentralized. In contrast to common approaches, one of the achievements of this project is to develop a control for UAVs able to assess a trade-off between localization accuracy and communication constraints.