Reproduce “RAD-LAD: Rule and Language Grounded Autonomous Driving in Real-Time” 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

Anurag Ghosh, Srinivasa Narasimhan, Manmohan Chandraker, Francesco Pittaluga

This research introduces RAD-LAD, a real-time hybrid autonomous driving system that integrates an enhanced rule-based planner with a language-grounded planner capable of real-time operation. The system demonstrates robust, explainable, and adaptive driving performance by combining dynamic geometric planning with fast language-based semantic reasoning for complex scenarios.

Key insights

  • A substantial portion of the 'difficulty' in existing long-tail benchmarks for autonomous driving stems from geometric limitations of rule-based planners rather than purely semantic challenges.
  • Language models can be adapted for real-time closed-loop autonomous driving by formulating planning as a single-pass classification over a trajectory vocabulary, effectively overcoming the latency bottleneck.
  • Language-based supervision provides a valuable inductive bias, enhancing the performance of learning-based planners and enabling explainable decision-making for complex and interactive driving scenarios.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.