Hi ODRI team!
We’re a team of undergraduate students interested in recreating and running experiments with the Solo 8. One of our project topics is using Reinforcement Learning (RL) to learn robot control, so we ended up creating an Open AI Gym API-compliant environment. The Open AI Gym API is the standard in the RL domain; so this Solo 8 environment has out-of-the-box compatibility with frameworks like Tensorflow.
You can find this simulator here: GitHub - WPI-MMR/gym_solo: A custom open ai gym environment for solo experimentation.
In theory, getting a fully interactive, position-controllable pybullet simulation of the solo 8 should be as easy as
git clone https://github.com/WPI-MMR/gym_solo.git && cd gym_solopip install .python examples/solo8_vanilla/interactive_pos_control.py
We also wanted to use this simulator to test out our hand-written controls as well. So, we’ve also implemented a realtime simulator wrapper that runs off of the real-time-clock and creates/listens to a virtual serial port. Then, you can connect to that port with any program/framework/language of your choice (our team has been using ROS) and interact with the simulation as if it was a real robot.
We really appreciate the effort that the ODRI team has put into the Solo 8–being able to build upon your guys’ work is the only reason that we’ve been able to pursue this project in our senior year. We wanted to help out the initiative in any way that we can and we feel like this is one way. We’ve designed this repo to to be fully open-source, with extendable environments, observations, rewards, etc, and healthy coding practices. While we can’t promise that we can work on issues (we need to graduate first!), we greatly appreciate any feedback.
We were wondering what was the best way to accredit your work–right now we just have a link and a description. However, we’re not sure that’s enough and want to make sure that you guys get every bit of credit that you deserve!