Flatland: a Lightweight First-Person 2-D Environment for Reinforcement Learning

Hugo Caselles-Dupré (1, 2) , Louis Annabi (1) , Oksana Hagen (1, 3) , Michael Garcia-Ortiz (1) , David Filliat (4, 2)

1 SoftBank Robotics Europe
2 Flowers - Flowing Epigenetic Robots and Systems
3 Plymouth University
4 U2IS - Unité d'Informatique et d'Ingénierie des Systèmes

Flatland is a simple, lightweight environment for fast prototyping and testing of reinforcement learning agents. It is of lower complexity compared to similar 3D platforms (e.g. Deep-Mind Lab or VizDoom), but emulates physical properties of the real world, such as continuity, multi-modal partially-observable states with first-person view and coherent physics. We propose to use it as an intermediary benchmark for problems related to Lifelong Learning. Flatland is highly customizable and offers a wide range of task difficulty to extensively evaluate the properties of artificial agents. We experiment with three reinforcement learning baseline agents and show that they can rapidly solve a navigation task in Flatland. A video of an agent acting in Flatland is available here: https://youtu.be/I5y6Y2ZypdA.

Domaines: Robotique [cs.RO]

Hugo Caselles-Dupré, Louis Annabi, Oksana Hagen, Michael Garcia-Ortiz, David Filliat. Flatland: a Lightweight First-Person 2-D Environment for Reinforcement Learning. Workshop on Continual Unsupervised Sensorimotor Learning, ICDL-EpiRob 2018, Sep 2018, Tokyo, Japan. ⟨hal-01951945⟩