Zero to Autonomy in Real-Time: Online Adaptation of Dynamics in Unstructured Environments

Abstract--Autonomous robots must go from zero prior knowledge to safe control within seconds to operate in unstructured environments. Abrupt terrain changes, such as a sudden transition to ice, create dynamics shifts that can destabilize planners unless the model adapts in real-time. We present a method for online adaptation that combines function encoders with recursive least squares, treating the function encoder coefficients as latent states updated from streaming odometry. We evaluate our approach on a V an der Pol system to highlight algorithmic behavior, in a Unity simulator for high-fidelity off-road navigation, and on a Clearpath Jackal robot, including on a challenging terrain at a local ice rink. Across these settings, our method improves model accuracy and downstream planning, reducing collisions compared to static and meta-learning baselines. High-speed ground vehicles require dynamics models that evolve as quickly as the terrain itself. When operating near the limits of controllability, even modest prediction errors in ground terrain interaction can lead to instability, skidding, or rollover. This problem is particularly apparent in off-road navigation: transitions such as pavement to loose gravel can change friction properties within seconds, while mixed terrain features introduce variation in the terrain properties that are difficult to accurately predict. Planning frameworks such as Model Predictive Path Integral Control (MPPI) [27] rely on an accurate model of the system dynamics to predict rollouts and select optimal control actions in real-time.