No announcement yet.

RoboSub 2015 Simulator and Free Google Cloud Credit / GPU time

  • Filter
  • Time
  • Show
Clear All
new posts

  • RoboSub 2015 Simulator and Free Google Cloud Credit / GPU time

    Hi Everyone,

    Former RoboSub competitor (and now full-time Deep Learning / Data Science nerd).

    Many teams in this competition have a nice ROS/VRML simulator to test their code (and even more have a nice war chest of video footage from previous competitions).
    To help my last two teams, I went a little bit "obsessive" working on a Three.js Simulator of the 2015 RoboSub competition.

    For those teams that do not have a good Simulator to produce synthetic images &/or would like to supplement their training data with additional synthetic data, I've decided to share with everyone some of that work in that hope that it useful to other teams.

    Demonstration Video I made (UCLA Students should contact me for Google Cloud Credit):

    Sample training Data:

    Simulator URL(s):

    VMRL files:

    2015 Obstacles .obj/.mtl files

    Transdec Geometry/Textures:

    Coleman 2016 Journal Paper (talks about Simulator):

    "SimBox 2015" Image Segmentation / Classification Dataset:

    Would be fascinated to know about it if anyone used the simulator tor train an A2C reinforcement learning agent successfully!

    Good luck in 2018 everyone!
    Last edited by ncantrell; 03-08-2018, 01:34 AM.

  • #2
    Awesome, thanks for sharing!