gym-ropod

Defines a Gazebo-based OpenAI gym environment for a ROPOD robot.

README contents

1. Defined environments
2. Gazebo entities
   1. Gazebo worlds
      • Square world
   2. Object models
3. Setup
4. Usage
5. Requirements
6. Acknowledgments

Defined environments

ropod-nav-discrete-v0

A navigation environment for a ropod with a discrete action space. In each episode, the robot's objective is to reach a randomly generated navigation goal without colliding with obstacles (a collision ends the episode).

The action space contains the following actions:

• 0: forward motion with 0.1m/s
• 1: left motion with 0.1m/s
• 2: right motion with 0.1m/s
• 3: left turn with 0.1m/s linear speed and 0.1m/s rotational speed
• 4: right turn with 0.1m/s linear speed and 0.1m/s rotational speed
• 5: backward motion with -0.1m/s

The observation at each step is a list of 500 laser measurements obtained from a front laser scanner.

The reward is calculated using the following equation:

Here, d is the distance from the robot to the goal, c_t indicates whether the robot has collided at time t, and a_t denotes the action taken at time t. We thus want the robot to reach the goal without making unnecessary direction changes and without collisions. The values of the constants c_1 and c_2 are set to -1000 and -10 respectively.

The step and reset environment functions return a tuple of four elements (goal, obs, reward, done), where:

• goal: a 2D pose in the format (x, y, theta) representing the current goal the robot is pursuing
• obs: a list of laser scans
• reward: the above reward
• done: a Boolean indicating whether the episode has finished

Gazebo entities

Gazebo worlds

Gazebo worlds are included under model_config/worlds. The world to be used in a simulation is specified when creating the environment and its elements are loaded on the fly. The currently defined worlds are briefly described below.

Square world

A simple square world without any static obstacles (other than the walls); any other obstacles are added on the fly at randomised positions. The world with the robot and two randomly added obstacles is shown below.

Object models

Object models used in the simulation are included under model_config/models (the robot model is however not in this repository, but in ropod_sim_model). These models are used for adding environment obstacles on the fly.

Setup

1. Set up the package:

(sudo) python setup.py [develop|install]

2. Set the ROPOD_GYM_MODEL_PATH environment variable to the absolute path of the model_config directory:

export ROPOD_GYM_MODEL_PATH=/path/to/gym-ropod/model_config

Usage

A simple usage example for the environment is given below. In this case, we load the square world and add five obstacles to it.

import gym
launch_file = '/path/to/my_simulation_launch_file.launch'

# create, render, and reset the environment
env = gym.make('ropod-nav-discrete-v0',
               launch_file_path=launch_file,
               env_type='square',
               number_of_obstacles=5)
env.render(mode='human')
env.reset()

# sample an action
action = env.action_space.sample()

# apply the sampled action and get information about the outcome
(goal, obs, reward, done) = env.step(action)

Test scripts that illustrate the environment use and should run out of the box can be found under test.

Requirements

• Python 3.5+
• gym
• numpy
• transforms3d
• shapely
• The ropod simulation
• rospkg
• termcolor

Acknowledgments

The implementation is heavily based on this Toyota HSR gym environment