Dream to Drive with Predictive Individual World Model
TLDR
Proposes a model-based RL method with a predictive individual world model for autonomous driving, capturing vehicle interactions and intentions via trajectory prediction.
Reasoning
The paper introduces a novel approach to model-based RL for autonomous driving by focusing on individual vehicle interactions and intentions, which addresses limitations of scene-level reconstruction. Strengths include a clear problem motivation and promising simulation results, but weaknesses include lack of real-world validation and potential scalability issues.
Read-first score
Read-first score 66.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.
Field roles
Rank sensitivity
Stability: volatile; rank range: 152.
Keyword Scores
Deep Analysis
Innovations
- Predictive Individual World Model (PIWM) that describes the driving environment from an individual-level perspective
- Captures vehicles' interactive relations and their intentions via a trajectory prediction task
- Joint learning of a behavior policy with PIWM, trained in the world model's imagination using intention-aware latent states
Methodology
The proposed method uses a predictive individual world model (PIWM) that models the driving environment at the individual vehicle level, capturing interactive relations and intentions through trajectory prediction. A behavior policy is learned jointly with PIWM and trained in the world model's imagination. The method is evaluated on simulation environments built from real-world challenging interactive scenarios.
Key Results
Compared to popular model-free and state-of-the-art model-based reinforcement learning methods, the proposed method achieves the best performance in terms of safety and efficiency.