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

arXiv 2026 50.3 method, system

TLDR

HOLO-MPPI combines offline high-level policy learning with online MPPI control using a learned world model for multi-scenario motion planning.

Reasoning

Strengths include a novel hierarchical framework that improves generalization across driving scenarios without per-scenario tuning. Weaknesses are the lack of real-world validation and domain specificity to autonomous driving.

Read-first score

Read-first score 50.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 24.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

94.8

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94782897

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,evaluation

Topical relevance 29%

34.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 396.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Combines high-level policy learning (offline) with low-level stochastic optimal control (online MPPI) for multi-scenario motion planning without per-scenario retuning.
High-level policy learned offline in an abstract action space with a learned world model serves as a data-driven prior generator that parameterizes MPPI's sampling distribution conditioned on current observation and goal.
Instantiation in autonomous driving with an effective high-level action space and tailored model architectures.

Methodology

Offline, a high-level policy is learned to propose scenario-robust plans in an abstract action space using a learned world model for online rollout. Online, this policy acts as a prior generator that parameterizes MPPI's sampling distribution conditioned on the current observation and goal; MPPI then optimizes low-level control sequences around this prior in real time to adapt to local disturbances. The framework is evaluated across diverse driving scenarios against MPPI and end-to-end RL baselines.

Key Results

HOLO-MPPI improves upon MPPI and end-to-end RL baselines while maintaining real-time control across diverse driving scenarios.

Limitations

Relies on a learned world model, which may be inaccurate in unseen or highly stochastic scenarios.
The high-level action space design may require manual tuning for different domains or tasks.
Real-time performance may degrade as scenario complexity increases due to the computational demands of MPPI and policy inference.
Evaluation is limited to autonomous driving; generalization to other robotics domains is not demonstrated.

Resources

Paper