Motion Planning

Project Overview

This project focuses on developing an efficient and robust motion planning system for autonomous vehicles (AVs) using reinforcement learning and other machine learning techniques. The system will enable the vehicle to navigate safely and efficiently through complex environments, considering obstacles, road structures, and dynamic traffic conditions. The motion planning algorithm will be trained using simulated driving scenarios(ISS) and evaluated for safety, efficiency, and adaptability to various real-world conditions.

Objectives

  • Simulation environment: use ISS, a high-fidelity simulation platform, to model diverse traffic, road conditions, and dynamic obstacles (e.g., pedestrians, other vehicles) for motion planning tasks.
  • Data collection: gather sensor data (e.g., LiDAR, cameras, radar), vehicle states, and control commands during simulation runs and real-world driving scenarios to train and validate the motion planning model.
  • Neural networks for motion planning: develop a neural network-based motion planning algorithm that can generate safe and efficient trajectories for AVs in complex environments.
  • Safety and efficiency evaluation: test and Verify the motion planning system across various scenarios to ensure safety (e.g., collision avoidance) and efficiency (e.g., optimal path generation, fuel consumption).

Deliverables

  • Motion planning algorithm: a neural network-based motion planning algorithm for autonomous vehicles capable of safely and efficiently navigating through dynamic environments.
  • Verification and validation: a comprehensive evaluation of the motion planning system’s performance using simulated, real-world driving scenarios and formal verification techniques.
  • Case studies: real-world simulations and case studies demonstrating the performance of the motion planning system in various traffic and road conditions.
  • Codebase: well-documented code and setup instructions for the motion planning system, including the reinforcement learning model and simulation environment.