Tutorials

In the tutorials, we first look at the Gymnasium and PettingZoo environments that integrate navground. Then, we use them to train navigation policies in different scenarios.

Our goal here is not to train the best possible policies or to perform a rigorous comparison between algorithms. In fact, we do not optimize hyper-parameters. Instead we want to show how navground.learning can be used to learn navigation policies in increasingly complex scenarios.

• Basics
  • Gymnasium Environment
    • Navground policy
  • Navground-PettingZoo integration
    • A single group
    • Two groups
    • Convert to a Gymnasium Env
    • Convert from a Gymnasium Env
    • Saving and loading
  • Using a ML policy in Navground
• Empty environment
  • Defining a scenario
  • Creating an enviroment
  • Training policies
    • Imitation Learning
    • Reinforcement Learning
  • Evaluating the policies
  • Exporting the trained policy
  • Using the policies in navground
• Corridor with obstacle
  • Scenario
    • Sensor
  • Learning
    • Imitation Learning with Behavior Cloning
    • Imitation learning with DAgger
    • Reinforcement learning with SAC
    • Comparision with HL
• Crossing
  • Training one agent among many agents
    • Scenario
    • Environment
    • Human-like behavior
    • Training
    • Summary
  • Performance of policies trained in single-agent environment
    • One agent following the policy
    • More agents following the policy
  • Training agents among peers
    • Environment
    • Pre-trained SA policies
    • Training
    • Summary
  • Performance of policies trained in multi-agent environment
    • Video
    • Reward
    • Final video
• Periodic Crossing
  • Uniform speeds
    • Scenario
    • Environment
    • Training
  • Different speeds
    • Scenario
    • Environment
    • Training
    • Varying target speed at runtime