The F1TENTH Gym environment

This is the repository of the F1TENTH Gym environment.

This project is still under heavy developement. Updated by UBM Driverless for Python 3.12 and replaced gym with gymnasium.

You can find the old documentation of the environment here.

Installation

To install the environment use pip install -e . in the root directory of the repository. It is advised to use a virtual environment for the installation if you just want to test the package.

Usage example

import numpy as np
import gymnasium as gym


params = {'mu': 1.0489,'C_Sf': 4.718,'C_Sr': 5.4562,'lf': 0.15875,'lr': 0.17145,'h': 0.074,'m': 3.74,'I': 0.04712,'s_min': -0.46,'s_max': 0.46,'sv_min': -3.2,'sv_max': 3.2,'v_switch': 7.319,'a_max': 9.51,'v_min':-5.0,'v_max': 20.0,'width': 0.31,'length': 0.58}
num_agents = 1


single_agent_low = np.array(
    [-np.inf, -np.inf, params['s_min'], params['v_min'], -np.pi, params['sv_min'], -np.inf], dtype=np.float32)
single_agent_high = np.array(
    [+np.inf, +np.inf, params['s_max'], params['v_max'], np.pi, params['sv_max'], +np.inf], dtype=np.float32)

# Duplicate for all agents to match the shape (num_agents, 7)
obs_low = np.tile(single_agent_low, (num_agents, 1))
obs_high = np.tile(single_agent_high, (num_agents, 1))

# Now create the observation space with the proper dimensions
observation_space = gym.spaces.Box(
    low=obs_low,
    high=obs_high,
    shape=(num_agents, 7),
    dtype=np.float32
)

single_agent_low = np.array([params['s_min'], 0.0], dtype=np.float32)
single_agent_high = np.array([params['s_max'], +np.inf], dtype=np.float32)

action_low = np.tile(single_agent_low, (num_agents, 1))
action_high = np.tile(single_agent_high, (num_agents, 1))

action_space = gym.spaces.Box(
    low=action_low,
    high=action_high,
    shape=(num_agents, 2),
    dtype=np.float32
)


def test_f110_env_rendering():
    # Initialize the environment
    map_filepath = ...  # Path to the map file
    raceline_filepath = ...  # Path to the raceline file
    
    env = gym.make('f110_gym:f110-v0',
                   map=map_filepath,
                   map_ext='.pgm',
                   num_agents=num_agents,
                   timestep=0.01,               # Simulation timestep
                   render_mode='human',         # None, 'human', 'human_fast'. If None, no rendering is done
                   points_in_foreground=False,  # Whether to render trail points of the car path in the foreground
                   raceline_path=raceline_filepath,
                   reward_function='s',         # 's', 's+d'
                   w_s=1.0,
                   w_d=0.3,
                   lower_bound_penalty_yaw_collision=1.0,
                   upper_bound_penalty_yaw_collision=2.0)

    # Set initial poses for agents
    # Format: [x, y, theta] for each agent
    # init_poses = np.array([
    #     [4.0, 5.0, 0.0],  # Agent 0 (ego)
    #     [0.0, 0.0, 0.0]  # Agent 1
    # ])
    init_poses = observation_space.sample()
    init_poses = init_poses[:, [0, 1, 4]]

    # Reset the environment with initial poses
    observation, info = env.reset(options={"poses": init_poses})
    print('obs reset:\n', observation)
    done = False

    # Run simulation for 1000 steps or until done
    for i in range(1000):
        if done:
            print(f"Episode finished after {i} steps")
            break

        # Generate random actions (steering angle, acceleration)
        # Actions format: [[steer_0, accel_0], [steer_1, accel_1], ...]
        # actions = np.array([
        #     [0, 0],
        #     [0, 0]
        # ])
        actions = action_space.sample()

        # Take action and get new observation
        observation, reward, terminated, truncated, info = env.step(actions)
        done = terminated or truncated

        # Print some information
        if i % 100 == 0:
            print(f"Step {i}")
            print(f"Ego position: ({observation[0,0]:.2f}, {observation[0,1]:.2f})")

        # Render the environment in human mode
        env.render()

if __name__ == "__main__":
    test_f110_env_rendering()

Citing

If you find this Gym environment useful, please consider citing:

@inproceedings{okelly2020f1tenth,
  title={F1TENTH: An Open-source Evaluation Environment for Continuous Control and Reinforcement Learning},
  author={O’Kelly, Matthew and Zheng, Hongrui and Karthik, Dhruv and Mangharam, Rahul},
  booktitle={NeurIPS 2019 Competition and Demonstration Track},
  pages={77--89},
  year={2020},
  organization={PMLR}
}