ForgeDrive: Bidirectional Cross-Conditioning for Unified Visual-Action Generation in Autonomous Driving

arXiv 2026 42.7 method, application

TLDR

ForgeDrive unifies visual and action generation in autonomous driving via bidirectional cross-conditioning, enabling act-then-imagine inference for improved planning and simulation.

Reasoning

The paper introduces a novel act-then-imagine paradigm that reduces error cascades from visual generation to action planning, and unifies multiple driving tasks in a single model. However, evaluation is limited to the NAVSIM simulator without real-world validation, and limitations are not discussed.

Read-first score

Read-first score 42.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 6%

100

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

Topical relevance 29%

77.1

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 408.

Keyword Scores

world model

video world model

generative world model

interactive world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified autoregressive diffusion framework with visual-action cross-conditioning for autonomous driving
Act-then-imagine paradigm replacing the traditional imagine-then-act pipeline to prevent error cascading
Factorization of future as per-timestep frame-action pairs with intertwined modalities
Decoupled diffusion timesteps and UniDiffuser-style noise scheduler enabling cross-modal inference
Inference paradigm where action generation does not require a clean future frame and generated action improves subsequent frame generation
Integration of future ego-status prediction to enhance planning
Single model unifying driving simulation, planning, and visual odometry

Methodology

ForgeDrive is an autoregressive diffusion model that generates sequences of frame-action pairs. Training decouples diffusion timesteps for visual and action modalities and uses a UniDiffuser-style noise scheduler to learn bidirectional cross-conditioning. Inference follows an act-then-imagine paradigm, generating actions first and then conditioning frame generation on them, with additional ego-status prediction.

Key Results

On NAVSIM, ForgeDrive outperforms existing strong planners without post-training and unifies driving simulation, planning, and visual odometry in a single model.

Resources

Paper