Particle-based methods have an important role to play in the prediction of off-road vehicle mobility, particularly in the transient states where the traditional semi-empirical models are known to be less accurate. However, it is well known that particle based methods come at the cost of much higher computational costs than the much simpler semi-empirical approaches. For real-time simulations that can often involve humans-in-the-loop, this can make it impossible to use particle-based methods. As a real-time alternative to semi-empirical models that are still able to capture transient effects captured by the particle methods, the authors propose a hybrid modelling approach which augments the semi-empirical methods by adding or subtracting additional force components that are determined by a machine learning algorithm. This algorithm is trained on data generated from particle-based simulation [1] in the hopes that the resulting hybrid model will better capture dynamic and transient effects without the full, added computational cost of particle-based approaches [2]. This work will summarize the development of the model, notably how a particle-based discrete element method approach is used to generate training data for the hybrid model. A test campaign in which an off-road vehicle undergoes a drawbar pull was carried out in order to demonstrate the kind of transient system state that the semi-empirical model has difficulty capturing accurately. In this test, gradually increasing wheel-slip is induced by applying a load to a hitch at the rear of the test vehicle until it is immobilized and the wheels are spinning out. As the wheel slip gradually increases, we can observe the slip-sinkage effect, in which the wheel sinks deeper into the soil as the wheels begin to slip. A comparison between the experiments, our proposed hybrid model and the traditional semi-empirical model will be presented to demonstrate the benefits of the hybrid modelling approach.