Set of custom environments used in the Inria Flowers team

Installation

git clone https://github.com/qdrn/gym_flowers.git
cd gym_flowers
pip install -e .

Then use it as follows:

import gym
import gym_flowers
env = gym.make('MultiTaskFetchArm4-v5')

List of supported environments

• MultiTaskFetchArm4-v5 Modification of the Fetch Arm environments from OpenAI Gym. A 7-DoF robotic arm faces 2 cubes. It has access to a set of 7 tasks. T_0: Place the gripper at the target location; T1: Place cube 1 at the 2D target location on the table; T_2: Place cube 1 at the 3D target location above cube 0; T3: Stack cube 0 above cube 1. Observations: list of all objects position and velocities. Actions: 3D cartesian actions and gripper action (4 in total). Reward 0 when goal is met (sparse reward), -1 otherwise.

• MultiTaskFetchArm8-v5 Adds 4 out-of-reach cubes corresponding to 4 extra distracting tasks.