Lightweight autonomous unmanned aerial vehicles (UAV) are emerging as a central component of a broad range of applications. However, autonomous navigation necessitates the implementation of perception algorithms, often deep neural networks (DNN), that process the input of sensor observations, such as that from cameras and LiDARs, for control logic. The complexity of such algorithms clashes with the severe constraints of these devices in terms of computing power, energy, memory, and execution time. In this paper, we propose NaviSplit, the first instance of a lightweight navigation framework embedding a distributed and dynamic multi-branched neural model. At its core is a DNN split at a compression point, resulting in two model parts: (1) the head model, that is executed at the vehicle, which partially processes and compacts perception from sensors; and (2) the tail model, that is executed at an interconnected compute-capable device, which processes the remainder of the compacted perception and infers navigation commands. Different from prior work, the NaviSplit framework includes a neural gate that dynamically selects a specific head model to minimize channel usage while efficiently supporting the navigation network. In our implementation, the perception model extracts a 2D depth map from a monocular RGB image captured by the drone using the robust simulator Microsoft AirSim. Our results demonstrate that the NaviSplit depth model achieves an extraction accuracy of 72.81 % while transmitting an extremely small amount of data (1.218 KB) to the edge server. When using the neural gate, as utilized by NaviSplit, we obtain a slightly higher navigation accuracy as compared to a larger static network by 0.3% while significantly reducing the data rate by 95%. To the best of our knowledge, this is the first exemplar of dynamic multi-branched model based on split DNNs for autonomous navigation.

LiDAR is rapidly emerging as a central sensor in many applications involving autonomous navigation. However, the execution of state-of-the-art neural models for the analysis of point clouds produced by LiDARs necessitates considerable computing power, energy and memory. As a consequence, real-time analysis – e.g., 3D object detection – on resource-constrained mobile platforms such as Unmanned Aerial Vehicles (UAV) is often impractical. In this paper, we evaluate the feasibility of real-time LiDAR-based object detection for UAVs using measures and data obtained from a real-world deployment. First, we demonstrate that state-of-the-art neural models for 3D object detection cannot be executed even in relatively powerful embedded computers suitable for airborne drones, such as the NVIDIA Jetson Nano. Then, we focus our attention on edge computing, where the UAV offloads the execution of the analysis model to a compute-capable device (an edge server) positioned at the network edge. The key challenge is that point clouds generated by LiDARs have a large size (1.2MB per point cloud frame). We evaluate the overall capture-to-output delay of a remote analysis loop experimentally for WiFi and using expected data rate for cellular LTE environments. Finally, we evaluate the performance of 3D object detection on available datasets for autonomous vehicles and emphasize the challenges posed by the ability of UAVs to move in the 3D space. With our LiDAR-UAV system, we achieved detections with averages of 86% accuracy, 71.6% precision, and 55.14% recall outputted with an average end-to-end delay of 1920ms.