Reproduce “Learning Multiple Probabilistic Decisions from Latent World Model in Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lingyu Xiao, Jiang-Jiang Liu, Sen Yang, Xiaofan Li, Xiaoqing Ye, Wankou Yang, Jingdong Wang
Researchers from Southeast University and Baidu Inc. developed LatentDriver, a framework that integrates a latent world model with a multiple probabilistic planner to enhance autonomous driving decision-making. This approach models both environmental uncertainty and action stochasticity, leading to improved arrival rates and reduced collision rates on the Waymax benchmark.
Key insights
- Effective autonomous driving requires modeling both environmental uncertainty and the multiple possible actions a vehicle could take.
- A latent world model can provide contextual information to a planner without requiring computationally expensive rollouts.
- Training a world model with actions generated by an intermediate layer of the planner, rather than ground truth actions, helps mitigate the self-delusion problem.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.