Manchester Stinger Motorsports AI

Our Objectives

  • Build a fully autonomous racing car compliant with Formula Student UK AI rules
  • Develop a simulation environment for testing
  • Create a future proof system for development over time

Autonomous Events

Teams

  • Build an instance segmentation dataset to train the neural network on the 4 cones types, humans and cars.
  • Train an instance segmentation network on the aforementioned dataset.
  • Sample the depths of the detected objects to build an accurate representation of the cone.
  • Train a general adversarial network to augment the weather conditions in the dataset to increase the size of the dataset.

  • Take in sensor information into an Extended Kalman Filter to estimate the position of the car.
  • Build up a map of the detected cones.
  • Using K-Means clustering, reduce noise in the detected cones.

  • Estimate the track boundaries the mapped track.
  • Calculate an optimal path for the car drive.
  • Build a dataset of tracks and optimal lines.

  • Calculate steering angle given the position of the car and upcoming path.
  • Calculate the speed given the position of the car and upcoming path.
  • Eventually, move over to a reinforcement learning approach.

  • Integrate as a node in a ROS 2 environment.
  • Interact with the vehicle via CAN-bus to control the car.
  • Log all testing data.
  • Visualise all data in real time.
  • Build a working simulator in gazebo that allows us to test codebase.
  • Build a reasonably accurate vehicle physics model.
  • Simulate CAN-bus interface.
  • Live visualisation of the vehicle in the environment.
  • Simulate vehicle sensors – GPS, IMU, Wheel Speed.