An accurate power loss prediction in the gearbox is desirable for improving vehicle efficiency. To achieve this objective, evaluating the power loss is necessary. However, power loss is influenced by factors such as the gearbox structure, operating conditions, and gear oil formulation, making power loss evaluation a bottleneck in practice. Therefore, a systematic modeling methodology was developed to evaluate the gearbox power loss in an E-Axle that focuses on the influence of the gear oil factors in the load and no-load cases. The gearbox used in a light-duty truck E-Axle was tested to verify the proposed model. The test was performed under various operating speeds, input loads, and oil temperatures, and four types of gear oil with different formulations were also included to quantify their influence on the power loss. The results showed that the gearbox power loss was significantly influenced by the E-Axle operating conditions, oil temperatures, and different gear oil formulations, promoting different power losses. The comparison results showed good consistency between the predicted power loss and the measured data. The proposed methodology can be utilized to effectively predict the power loss of the E-Axle gearbox and further improve the E-Axle efficiency by selecting suitable oil formulations and adjusting oil temperatures.

Regular lubricating oil change in the gearbox is desirable for improving vehicle reliability and reducing operating costs. To achieve this objective, evaluating the oil change interval is necessary. However, due to the complex and dynamic properties of oil degradation, oil change interval evaluation has been a bottleneck in practice. Therefore, a solution strategy is proposed in this paper that utilizes the oil physicochemical properties derived from oil analysis data to determine the optimal oil change interval. With a large amount of oil analysis data collected, the iron (Fe) debris, kinematic viscosity (100 °C), and total acid number (TAN) are considered to be the oil change indicators of lubricating oil. By monitoring the changes in the selected oil change indicators, linear regression is firstly applied to the original oil analysis data to reveal the dynamic degradation process. Then, the Wiener-based stochastic process is used to describe the first hitting time and the increasing trends of the selected oil change indicator. Finally, the oil change interval can be obtained under the concept of the first hitting time. Compared with the planned maintenance time, the proposed method seems reasonable considering the dynamic property of oil degradation. The effectiveness of the proposed method is evaluated using a case study with an oil analysis dataset from an E-axle with a two-shift gearbox. The results show that the oil change interval increased by approximately 10,000 km (50%) compared with the planned maintenance interval. This will reduce vehicle maintenance time and save maintenance costs.

Trajectory planning is a central problem in autonomous driving, requiring long-horizon reasoning, strict safety guarantees, and robustness to rare but critical events. Recent learning-based planners increasingly formulate planning as an autoregressive sequence generation problem, analogous to large language models, where future motions are discretized into action tokens and predicted by Transformer-based neural sequence models. Despite promising empirical results, most existing approaches adopt time-domain action representations, in which consecutive actions are highly correlated. When combined with autoregressive decoding, this design induces degenerate generation behavior in learning-based planners, encouraging local action continuation and leading to rapid error accumulation during closed-loop execution, particularly in safety-critical corner cases such as sudden pedestrian emergence. To address this limitation of time-domain autoregressive planning, we propose a unified trajectory planning framework built upon three core ideas: (1) explicit action tokenization for long-horizon planning, (2) transformation of the action space from the time domain to the frequency domain, and (3) a hybrid learning paradigm that combines imitation learning with reinforcement learning. By representing future motion using compact frequency-domain action coefficients rather than per-timestep actions, the proposed planner is encouraged to reason about global motion intent before refining local details. This change in action representation fundamentally alters the inductive bias of learning-based autoregressive planning, mitigates exposure bias, and enables earlier and more decisive responses in complex and safety-critical environments. We present the model formulation, learning objectives, and training strategy, and outline a comprehensive experimental protocol.