Research On Energy-Efficiency Optimization Of Unmanned Autonomous Vehicle

Unmanned Aerial Vehicle（UAV）is widely used in daily life due to its high flexibility.However,as application scenarios become more complex,UAVs need to consume more power to support their missions.On the one hand,as a power source,the UAV’s motors consume a lot of energy to maintain or change its attitude.On the other hand,UAVs also consume a lot of energy to gather information and make decisions in order to keep the flight path correct.However,as an energy-constrained mobile device,the limited battery capacity that the UAV can carry means that the UAV has limited endurance,which indirectly leads to reduced mission success rate.Therefore,the energy-efficiency of UAV is a very important issue in its development.This paper focuses on the energy-efficiency of UAVs and considers its optimization from the perspectives of both software and hardware.At the software level,from the perspective of algorithm,we quantitatively analyze the sensing data and decision between the two adjacent steps of autonomous UAV based on DQN（Deep Q-Network）algorithm,and find that there are sensing consistency and Q-values symmetry.Based on sensing consistency,we propose the sensing approximation technique AERO-S.According to the symmetry of Q-values,we propose the decision approximation technique AERO-AP.Further,by combining the two technologies,we propose the approximate technology AERO.At the hardware level,from the perspective of processor,we experimentally demonstrate the existence of voltage margin in the chip design process of GPU as an example,and analyze the causes of voltage noise of processor from the perspective of microarchitecture activities.In view of the large margin between the supply voltage of the chip and the minimum voltage to ensure the correct execution of programs,we propose the energy-efficiency optimization of GPU based on voltage margin reduction.In this paper,the above two technologies are evaluated from the perspectives of mission success rate,energy consumption or power consumption.We evaluated AERO in four different environments.The results showed that AERO achieved significant energy-efficiency optimization with almost no cost of success rate,especially in an obstacle-free environment,which achieved about 25% RIT（relative improvement of time）and 19% RIE（relative improvement of energy）.As for the evaluation of the energy-efficiency optimization technology of GPU based on voltage margin reduction,it reduces dynamic power consumption by 32% and static power consumption by 18%on average.