Reproduce “ReasonNet: End-to-End Driving with Temporal and Global Reasoning” 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
Hao Shao, Letian Wang, Ruobing Chen, Steven L. Waslander, Hongsheng Li, Yu Liu
The ReasonNet framework enhances end-to-end autonomous driving systems by integrating explicit temporal and global reasoning modules. It achieved the top rank on the CARLA Leaderboard with a driving score of 79.95 and demonstrated improved performance in complex urban occlusion scenarios, reducing collision rates and improving infraction scores.
Key insights
- Explicitly fusing temporal information via a memory-augmented reasoning module significantly improves motion estimation and tracking of intermittently occluded objects, demonstrating up to a 46% increase in success rate on occlusion benchmarks.
- Modeling interactions between objects and the environment through global reasoning enables anticipation of potential dangers from constantly occluded objects by inferring contextual cues.
- A dedicated benchmark (DriveOcclusionSim) for various occlusion scenarios is crucial for systematically evaluating and advancing the robustness of autonomous driving systems in challenging, long-tail events.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.