Browsing by Author "Dressler, Falko"
Now showing 1 - 7 of 7
Results Per Page
Sort Options
- Continuous Maneuver Control and Data Capture Scheduling of Autonomous Drone in Wireless Sensor NetworksPublication . Li, Kai; Ni, Wei; Dressler, FalkoThanks to flexible deployment and excellent maneuverability, autonomous drones are regarded as an effective means to enable aerial data capture in large-scale wireless sensor networks with limited to no cellular infrastructure, e.g., smart farming in a remote area. A key challenge in drone-assisted sensor networks is that the autonomous drone's maneuvering can give rise to buffer overflows at the ground sensors and unsuccessful data collection due to lossy airborne channels. In this paper, we propose a new Deep Deterministic Policy Gradient based Maneuver Control (DDPG-MC) scheme which minimizes the overall data packet loss through online training instantaneous headings and patrol velocities of the drone, and the selection of the ground sensors for data collection in a continuous action space. Moreover, the maneuver control of the drone and communication schedule is formulated as an absorbing Markov chain, where network states consist of battery energy levels, data queue backlogs, timestamps of the data collection, and channel conditions between the ground sensors and the drone. An experience replay memory is utilized onboard at the drone to store the training experiences of the maneuver control and communication schedule at each time step.
- Data-driven Deep Reinforcement Learning for Online Flight Resource Allocation in UAVaided Wireless Powered Sensor NetworksPublication . Li, Kai; Ni, Wei; Kurunathan, Harrison; Dressler, FalkoIn wireless powered sensor networks (WPSN), data of ground sensors can be collected or relayed by an unmanned aerial vehicle (UAV) while the battery of the ground sensor can be charged via wireless power transfer. A key challenge of resource allocation in UAV-aided WPSN is to prevent battery drainage and buffer overflow of the ground sensors in the presence of highly dynamic lossy airborne channels which can result in packet reception errors. Moreover, state and action spaces of the resource allocation problem are large, which is hardly explored online. To address the challenges, a new data-driven deep reinforcement learning framework, DDRL-RA, is proposed to train flight resource allocation online so that the data packet loss is minimized. Due to time-varying airborne channels, DDRL-RA firstly leverages long short-term memory (LSTM) with pre-collected offline datasets for channel randomness predictions. Then, Deep Deterministic Policy Gradient (DDPG) is studied to control the flight trajectory of the UAV, and schedule the ground sensor to transmit data and harvest energy. To evaluate the performance of DDRL-RA, a UAV-ground sensor testbed is built, where real-world datasets of channel gains are collected. DDRL-RA is implemented on Tensorflow, and numerical results show that DDRL-RA achieves 19\% lower packet loss than other learning-based frameworks.
- Data-driven Flight Control of Internet-of- Drones for Sensor Data Aggregation using Multi-agent Deep Reinforcement LearningPublication . Li, Kai; Ni, Wei; Emami, Yousef; Dressler, FalkoEnergy-harvesting-powered sensors are increasingly deployed beyond the reach of terrestrial gateways, where there is often no persistent power supply. Making use of the internet of drones (IoD) for data aggregation in such environments is a promising paradigm to enhance network scalability and connectivity. The flexibility of IoD and favorable line-of-sight connections between the drones and ground nodes are exploited to improve data reception at the drones. In this article, we discuss the challenges of online flight control of IoD, where data-driven neural networks can be tailored to design the trajectories and patrol speeds of the drones and their communication schedules, preventing buffer overflows at the ground nodes. In a small-scale IoD, a multi-agent deep reinforcement learning can be developed with long short-term memory to train the continuous flight control of IoD and data aggregation scheduling, where a joint action is generated for IoD via sharing the flight control decisions among the drones. In a large-scale IoD, sharing the flight control decisions in real-time can result in communication overheads and interference. In this case, deep reinforcement learning can be trained with the second-hand visiting experiences, where the drones learn the actions of each other based on historical scheduling records maintained at the ground nodes.
- Deep Reinforcement Learning for Persistent Cruise Control in UAV-aided Data CollectionPublication . Kurunathan, John Harrison; Li, Kai; Ni, Wei; Tovar, Eduardo; Dressler, FalkoAutonomous UAV cruising is gaining attention dueto its flexible deployment in remote sensing, surveillance, andreconnaissance. A critical challenge in data collection with theautonomous UAV is the buffer overflows at the ground sensorsand packet loss due to lossy airborne channels. Trajectoryplanning of the UAV is vital to alleviate buffer overflows as wellas channel fading. In this work, we propose a Deep DeterministicPolicy Gradient based Cruise Control (DDPG-CC) to reducethe overall packet loss through online training of headings andcruise velocity of the UAV, as well as the selection of the groundsensors for data collection. Preliminary performance evaluationdemonstrates that DDPG-CC reduces the packet loss rate byunder 5% when sufficient training is provided to the UAV.
- Leverage variational graph representation for model poisoning on federated learningPublication . Li, Kai; Yuan, Xin; Zheng, Jingjing; Ni, Wei; Dressler, Falko; Jamalipour, AbbasThis article puts forth a new training data-untethered model poisoning (MP) attack on federated learning (FL). The new MP attack extends an adversarial variational graph autoencoder (VGAE) to create malicious local models based solely on the benign local models overheard without any access to the training data of FL. Such an advancement leads to the VGAE-MP attack that is not only efficacious but also remains elusive to detection. VGAE-MP attack extracts graph structural correlations among the benign local models and the training data features, adversarially regenerates the graph structure, and generates malicious local models using the adversarial graph structure and benign models’ features. Moreover, a new attacking algorithm is presented to train the malicious local models using VGAE and sub-gradient descent, while enabling an optimal selection of the benign local models for training the VGAE. Experiments demonstrate a gradual drop in FL accuracy under the proposed VGAE-MP attack and the ineffectiveness of existing defense mechanisms in detecting the attack, posing a severe threat to FL.
- LSTM-characterized Deep Reinforcement Learning for Continuous Flight Control and Resource Allocation in UAV-assisted Sensor NetworkPublication . Li, Kai; Ni, Wei; Dressler, FalkoUnmanned aerial vehicles (UAVs) can be employed to collect sensory data in remote wireless sensor networks (WSN). Due to UAV's maneuvering, scheduling a sensor device to transmit data can overflow data buffers of the unscheduled ground devices. Moreover, lossy airborne channels can result in packet reception errors at the scheduled sensor. This paper proposes a new deep reinforcement learning based flight resource allocation framework (DeFRA) to minimize the overall data packet loss in a continuous action space. DeFRA is based on Deep Deterministic Policy Gradient (DDPG), optimally controls instantaneous headings and speeds of the UAV, and selects the ground device for data collection. Furthermore, a state characterization layer, leveraging long short-term memory (LSTM), is developed to predict network dynamics, resulting from time-varying airborne channels and energy arrivals at the ground devices. To validate the effectiveness of DeFRA, experimental data collected from a real-world UAV testbed and energy harvesting WSN are utilized to train the actions of the UAV. Numerical results demonstrate that the proposed DeFRA achieves a fast convergence while reducing the packet loss by over 15%, as compared to existing deep reinforcement learning solutions.
- When Internet of Things meets Metaverse: Convergence of Physical and Cyber WorldsPublication . Li, Kai; Cui, Yingping; Li, Weicai; Lv, Tiejun; Yuan, Xin; Li, Shenghong; Ni, Wei; Simsek, Meryem; Dressler, FalkoIn recent years, the Internet of Things (IoT) is studied in the context of the Metaverse to provide users immersive cyber-virtual experiences in mixed reality environments. This survey introduces six typical IoT applications in the Metaverse, including collaborative healthcare, education, smart city, entertainment, real estate, and socialization. In the IoT-inspired Metaverse, we also comprehensively survey four pillar technologies that enable augmented reality (AR) and virtual reality (VR), namely, responsible artificial intelligence (AI), high-speed data communications, cost-effective mobile edge computing (MEC), and digital twins. According to the physical-world demands, we outline the current industrial efforts and seven key requirements for building the IoT-inspired Metaverse: immersion, variety, economy, civility, interactivity, authenticity, and independence. In addition, this survey describes the open issues in the IoT-inspired Metaverse, which need to be addressed to eventually achieve the convergence of physical and cyber worlds.