HOLO-MPPI: Multi-Scenario Motion Planning via Hierarchical Policy Optimization

HOLO-MPPI introduces a multi-scenario motion planning framework that integrates high-level policy learning with low-level stochastic optimal control, specifically designed for autonomous driving applications. The approach involves an offline learned high-level policy that generates robust plans in an abstract action space, which then informs a model predictive path integral (MPPI) control mechanism for real-time optimization. Evaluations indicate that HOLO-MPPI outperforms traditional MPPI and end-to-end reinforcement learning baselines, making it a significant advancement for practitioners aiming to enhance the adaptability and robustness of motion planning in dynamic environments.