MDP Modelling for Motion Planning in Autonomous Racing Cars

Project Overview

Autonomous racing cars require advanced motion planning techniques to navigate high-speed environments. Efficient and safe motion planning for autonomous racing cars demands real-time decision-making that optimizes vehicle performance while accounting for track geometry, vehicle dynamics, and safety constraints. This project proposes the use of Markov Decision Processes (MDPs) to model the motion planning problem in autonomous racing, aiming to design a system capable of making optimal decisions at each step of a race.

Objectives

The main objective of this project is to apply MDPs to autonomously plan the motion of a racing car. The system should be able to:

  • navigate the track with high efficiency, optimizing speed and trajectory;
  • respond dynamically to the racing environment, including varying track conditions and unforeseen obstacles; and
  • balance between safety (avoiding collisions) and performance (maximizing speed).

MDP Framework for Motion Planning

In the context of motion planning, an MDP provides a formal framework for modeling sequential decision-making. The key elements of an MDP include:

  • States: a representation of the car’s current situation (e.g., position, velocity, heading).
  • Actions: the set of possible actions the car can take (e.g., steering, acceleration, braking).
  • Transition model: a model that predicts how the car’s state evolves when an action is taken.
  • Reward function: a mechanism that assigns rewards or penalties based on the desirability of a given state-action pair.
  • Policy: a strategy that defines the best action to take in each state to maximize the cumulative reward.

Expected Outcomes

This project will result in an MDP-based motion planner that:

  • can navigate a racing track in a manner that optimizes speed while ensuring safety;
  • makes dynamic decisions based on changing conditions in the racing environment; and
  • demonstrates adaptability to various types of tracks and racing scenarios.

The outcome will be validated through simulations, comparing the MDP-based planner’s performance against traditional motion planning approaches.