Mentored Research ProjectsRepresentative program

Talking to Self-Driving Cars: Human Intent and Decision-Making with LLMs

Study how natural-language preferences can be interpreted and evaluated in a simulated driving decision system.

Overview

Learners build a bounded simulation linking a human instruction, a language-model interpretation, and a constrained driving decision. The research focuses on intent, ambiguity, feedback, and safety boundaries. All proposed testing is simulated; the project does not authorize real-vehicle or public-road experimentation.

Learning goals and possible work

Proposed learning outcomes for this program example:

  • Define a constrained language-to-decision research question.
  • Build a simulated interaction pipeline and evaluation cases.
  • Analyze failures, ambiguity, and limits of model behavior.

Illustrative learning sequence

The sequence below illustrates how this program’s content can be organized. Topics, pacing, and project depth are adapted for each offering. This is not an archived record of a specific cohort’s weekly syllabus.

  1. 01Human intent and a bounded driving-simulation problem
  2. 02Driving decisions, constraints, and evaluation scenarios
  3. 03Representing natural-language instructions and ambiguity
  4. 04Language-model interpretation and structured outputs
  5. 05Connecting intent to a simulated decision policy
  6. 06Human-in-the-loop feedback and interface design
  7. 07Testing instructions, variations, and failure cases
  8. 08Safety constraints and robustness in simulation
  9. 09Analyzing results and explaining system limitations
  10. 10Simulated demonstration and research report
Ask About This Program Back to the Program Portfolio

Related program examples