Comparing Deep Learning Models for Battery State Prediction based on Robot Path Plans
Contributing USMA Research Unit(s)
Electrical Engineering and Computer Science, Robotics Research Center
IEEE MIT Undergraduate Research Technology Conference (URTC)
This research compared the accuracy of two deep learning models (i.e., neural network based regression and recurrent neural network) for predicting the state of a robot’s lithium-ion battery during waypoint navigation. Unlike previous research, the models in this paper were trained using real-world data collected from a robot that operated under realistic conditions and in a variety of terrain types. Robotic systems that can accurately predict the amount of energy required to navigate between waypoints can determine the feasibility of a route prior to starting or compare route options for better efficiency. Experimental results show the accuracies of the prediction models increase as the robot approaches its objective. Furthermore, the aggregation of the data reveals that there is a 4.64% statistical difference between the accuracy of the two models with the recurrent neural network (RNN) performing the best.
DeVries, A.L., Moore, J.T., Miller, S.J., Lowrance, C.J., “Comparing Deep Learning Models for Battery State Prediction based on Robot Path Plans”, IEEE MIT Undergraduate Research Technology Conference (URTC), Oct. 5-7, Cambridge, MA, 2018.
Record links to items hosted by external providers may require fee for full-text.