The goal of this NSF CPS research project is to achieve true onboard autonomy in real time for UAVs in the absence of remote control and external navigation aids. Very low power and light weight machine intelligence techniques will be investigated to achieve multi-modal sensing, onboard detection, and adaptive control. Detection, optimization and control problems in an autonomous UAV will be formulated and solved using deep neural networks (DNN) and deep reinforcement learning (DRL). Ultra-low power and high-performance DNNs using the circulant weight matrix and FFT/IFFT operations will be trained and implemented on either GPUs or FPGAs. This unique technique has the potential to reduce computational complexity as well as the storage complexity of the DNNs, and hence enable us to close the loop of sensing, detection, and control in real-time.
Coming Soon.