TrafficBots: Towards World Models for Autonomous Driving Simulation and Motion Prediction

ICRA 23 2023 70 method, application

TLDR

TrafficBots formulates data-driven traffic simulation as a world model for autonomous driving, enabling configurable and scalable multi-agent simulation and motion prediction.

Reasoning

The paper presents a novel world model for autonomous driving simulation, addressing configurability via destination and personality, and scalability via new positional encoding. Strengths include clear formulation and practical experiments on Waymo dataset. Weaknesses include limited evaluation to motion prediction and lack of full generative or interactive capabilities.

Read-first score

Read-first score 70, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=dataset,github

Methodology quality 25%

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

Recency 8%

65.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2023

Topical relevance 42%

64.3

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

Field roles

Methodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 147.

Keyword Scores

world model

world simulator

world dynamics prediction

generative world model

interactive world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Formulating data-driven traffic simulation as a world model tailored for autonomous driving planning
Introducing a destination and a time-invariant latent personality for each agent to enable configurable behaviors
A new positional encoding scheme for angles that allows all agents to share the same vectorized context and use dot-product attention, improving scalability

Methodology

TrafficBots is a multi-agent policy built upon motion prediction and end-to-end driving. It uses a destination and a latent personality per agent for configurable behavior, and a novel positional encoding for angles to enable shared vectorized context and dot-product attention. The model is trained and evaluated on the Waymo open motion dataset.

Key Results

TrafficBots simulates realistic multi-agent behaviors and achieves good performance on the motion prediction task on the Waymo open motion dataset.

Resources

Paper Code