Learning Automata Models from ISS

Contact: Yong Li

Project Overview

This project aims to learn an automaton model (e.g., a finite state machine or an omega automaton) for autonomous driving using data from our ISS platform. The model will represent high-level decision-making logic for autonomous vehicles (AVs), such as transitioning between driving behaviors (e.g., breaking, stopping at intersections). The automaton model will be learned from simulated driving scenarios and then evaluated for safety and efficiency.

Objectives

• Simulation environment: use ISS to simulate diverse driving scenarios (traffic, weather, road conditions).
• Data collection: gather sensor data (LiDAR, cameras, radar), vehicle states, and control commands during the simulation.
• Automaton model learning: apply automata learning algorithms like active learning and passive learning to extract state-transition rules that model the vehicle’s decision-making process.
• Evaluation: test the learned model on new scenarios to measure safety, efficiency, and robustness. Validate the model’s performance in real-world simulations.
• Evaluation metrics: measure safety (collision rate, traffic law compliance).

Deliverables

• Automaton model: a state-based decision-making model for AVs.
• Simulation integration: a working pipeline for generating and learning from simulation data.
• Codebase: well-documented code and setup instructions.