Reproduce “Affordance Learning In Direct Perception for 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
Chen Sun, Jean M. Uwabeza Vianney, Dongpu Cao
Researchers from the University of Waterloo developed a cost-effective method for training autonomous driving perception models by automatically generating labeled training data from Google Street View images and OpenStreetMap data. This approach trains neural networks to recognize driving affordances, achieving performance comparable to or exceeding human baselines and demonstrating generalization across diverse geographical regions.
Key insights
- Cost-effective training data for autonomous driving perception can be generated automatically by leveraging existing geospatial data like Google Street View and OpenStreetMap.
- Training deep learning models on a diverse set of automatically labeled data from different geographical regions enhances cross-regional generalization.
- A custom convolutional neural network can effectively learn driving affordances from automatically labeled visual data, achieving high performance with reduced data volume compared to prior work.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.