Reproduce “Multi-task Learning with Attention for End-to-end 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
Keishi Ishihara, Anssi Kanervisto, Jun Miura, Ville Hautamäki
The authors train a driving network to predict controls while jointly learning road segmentation, depth, and traffic-light states, using task-specific attention to route visual features. In CARLA 0.8.4, it reaches 37 ± 9% success versus 23 ± 5% without attention in New town, Dense NoCrash, and records 4.2 ± 3.3% red-light crossings versus 14.8 ± 8.1% for One Turn in New town and weather on CoRL2017; gains are condition-specific.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.