Characterization of Load Uncertainty in Unstructured Terrains and Applications to Battery Remaining Run‐time Prediction

Deployment of field robots in unstructured environments continues to rise, and the electrochemical battery remains the de facto standard for energy storage in robotic applications. However, robot mission planning, which relies on battery depletion time information, enforces conservative operation due to the lack of statistical rigor on the run‐time predictions. A two‐tier self‐supervised load characterization methodology for mobile robots operating in unstructured environments is proposed. Coupled with load characterization, a model‐based statistical battery remaining run‐time prediction algorithm utilizing particle filtering is presented. Given measured power loads during operation, the characterization algorithm employs Gaussian mixture modeling to cluster measured loads into a priori unknown regions. With clustered power demand regions, the computation of transition probabilities between the mixture models provides a jump Markov characterization of the historical power loads. An experimental study utilized Packbot data gathered during operation on general desert terrain. A particle filter prediction framework was shown to more accurately predict the remaining run‐time of the Packbot given the unstructured terrain compared to existing load‐averaging techniques.