Map-World: Masked Action planning and Path-Integral World Model for Autonomous Driving

2025-11-25

AI Summary

Autonomous driving multimodal motion planning faces dual challenges in modeling multiple future scenarios and ensuring computational efficiency. Existing approaches rely on handcrafted anchors or reinforcement learning to select a single mode, leading to information loss and complex optimization. This paper proposes an anchor-free Masked Action Planning (MAP) framework: it pioneers the application of masked sequence completion to driving planning, coupled with latent variable expansion to generate diverse trajectory queries; introduces a path-weighted world model that jointly models scene dynamics and discrete path-integration weights in the bird’s-eye view (BEV) space, enabling end-to-end optimization of multimodal semantic loss; and enhances trajectory diversity via noise injection and latent-space state expansion. Evaluated on NAVSEM, MAP achieves state-of-the-art performance among world-model-based methods, matches anchor-based approaches in accuracy, supports real-time inference, and eliminates the need for reinforcement learning.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Multiagent Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Motion planning for autonomous driving must handle multiple plausible futures while remaining computationally efficient. Recent end-to-end systems and world-model-based planners predict rich multi-modal trajectories, but typically rely on handcrafted anchors or reinforcement learning to select a single best mode for training and control. This selection discards information about alternative futures and complicates optimization. We propose MAP-World, a prior-free multi-modal planning framework that couples masked action planning with a path-weighted world model. The Masked Action Planning (MAP) module treats future ego motion as masked sequence completion: past waypoints are encoded as visible tokens, future waypoints are represented as mask tokens, and a driving-intent path provides a coarse scaffold. A compact latent planning state is expanded into multiple trajectory queries with injected noise, yielding diverse, temporally consistent modes without anchor libraries or teacher policies. A lightweight world model then rolls out future BEV semantics conditioned on each candidate trajectory. During training, semantic losses are computed as an expectation over modes, using trajectory probabilities as discrete path weights, so the planner learns from the full distribution of plausible futures instead of a single selected path. On NAVSIM, our method matches anchor-based approaches and achieves state-of-the-art performance among world-model-based methods, while avoiding reinforcement learning and maintaining real-time inference latency.

Problem

• Handles multiple plausible futures in autonomous driving motion planning
• Eliminates reliance on handcrafted anchors or reinforcement learning
• Learns from full distribution of futures instead of single path

Innovation

• Masked action planning treats future motion as sequence completion
• Compact latent state expands into diverse trajectory queries with noise
• Lightweight world model rolls out BEV semantics for candidate trajectories