Enabling Multimodal Sensing, Real-time Onboard Detection and Adaptive Control for Fully Autonomous Unmanned Aerial Systems

ABSTRACT

Autonomous systems such as unmanned aerial vehicles (UAV) have gradually gaining importance in daily life. The global market for commercial applications of the UAV technology will rise to as much as $127 billion by 2020 with more than 6,000% increase by the end of the decade. The goal of this proposed research project is to achieve true onboard autonomy in real time for small UAVs in the absence of remote control and external navigation aids. Very low power and lightweight machine intelligence techniques will be investigated to achieve multi-modal sensing, onboard object detection/classification, and adaptive control. We will formulate and solve the detection, optimization and control problems using deep neural networks (DNN) and deep reinforcement learning (DRL).

The proposed project will formulate and solve the detection and control problems in different layers of an autonomous UAV using DNNs. The unified computation model simplifies the hardware design. Our block circulant-matrix based DNN allows very low-cost, low-power and highly optimized hardware. The goal is to close the control loop in real-time by utilizing our unique technique in realizing the DNN.