Autonomous heavy-duty vehicles (HDVs) such as trucks and trailers are poised to transform logistics, mobility, and large-scale industrial operations. Yet, reliable real-world deployment remains constrained by the difficulty of achieving safe motion planning under diverse and adverse weather conditions, especially in northern countries such as Sweden. Rain, snow, fog, low temperatures, and poor visibility can degrade perception quality and modify the physical interaction between the vehicle and the road—particularly through changes in surface friction. These effects increase uncertainty in both environmental understanding and vehicle dynamics, making robust motion prediction and trajectory planning especially challenging for large, high-inertia HDVs. Conventional planning frameworks struggle in such scenarios because they rely on rigid models that do not adapt to weather-driven variability. Our project, Neural Motion Planning for Autonomous Heavy-Duty Vehicles in All-Weather Conditions, aims to develop next-generation AI-driven planning systems capable of maintaining safety, efficiency, and stability across a full spectrum of environmental disturbances. A central component of the research is the integration of neural motion prediction models that anticipate the future behavior of surrounding agents and environmental evolution even when uncertainty is high. To enhance physical realism and robustness, the project incorporates friction estimation modules—learned directly from multimodal sensor data—that infer road–tire interaction properties under varying weather. These friction estimates are used both in predicting the behavior of other agents (such as vehicles, pedestrians, or other road users) and in trajectory planning through adaptive speed control, braking margins, and maneuvering constraints. By coupling friction-aware dynamics with learned representations of the environment, the system is capable of generating more accurate, risk-sensitive trajectories that respect the unique operational characteristics of heavy-duty vehicles. The project further explores neural trajectory planners trained using large-scale multimodal datasets and simulated environments. These planners are exposed to a wide range of weather conditions and road surfaces through real-world data and high-fidelity, domain-randomized simulation. Weather augmentation, friction variation, and sensor degradation modeling are used to improve generalization and robustness. While the primary focus is on modular architectures that integrate prediction, friction estimation, and planning, the project is also aiming to extend these efforts toward end-to-end learning pipelines, where perception, prediction, and control are jointly optimized in a unified neural framework. This direction is expected to further increase adaptability and system resilience in complex all-weather conditions. The anticipated outcome is a scalable, weather-resilient neural motion planning framework capable of improving the safety and operational efficiency of autonomous HDVs across ports, mining sites, industrial zones, and public road networks. Achieving these goals requires compute resources for training large neural models, performing friction-aware driving simulation at scale, and conducting multi-scenario evaluation campaigns. The proposed research aims to significantly advance the state of all-weather autonomous mobility and accelerate the real-world deployment of safe and robust heavy-duty vehicle autonomy.