<![CDATA[The deployment of autonomous mobile robots in smart manufacturing and intralogistics has grown rapidly, yet current battery management systems can only monitor real-time charge levels without predicting future energy consumption. This reactive limitation risks mid-mission battery depletion and production disruption. The study presents a predictive analytics system for the Polebot autonomous mobile robot integrated with a Robot Operating System 2 data historian pipeline and an InfluxDB time-series database. The objective is to evaluate whether an autoregressive time-series model or a gradient-boosted machine learning model better suits different operational conditions, specifically constant-velocity static operation versus acceleration-heavy dynamic operation. Data were collected from three sensor sources across six operational protocols covering baseline, high-load, stop-and-go, creep, burst acceleration, and mixed conditions, yielding 10,800 synchronized data points at 1 Hz after resampling. Results show that the Autoregressive Integrated Moving Average model with parameters (2,1,3) achieves a Mean Absolute Error of 1.047% and a symmetric Mean Absolute Percentage Error of 1.74% for battery State of Charge prediction under static conditions. Extreme Gradient Boosting achieves a Mean Absolute Error of 0.022 watts for motor power prediction, 136 times more accurate than the time-series model for the same variable. The proposed Condition-Based Temporal Switching framework was validated on 1,803 data points and autonomously produced 265 model transitions during a 30-minute mixed operational test, with static conditions comprising 83.9% of the validation window. Adaptive model selection outperforms single-model strategies for energy prediction in autonomous mobile robot platforms.]]>
