Skip to content

Latest commit

 

History

History
8 lines (7 loc) · 685 Bytes

README.md

File metadata and controls

8 lines (7 loc) · 685 Bytes

ball-balancer

  • Demonstration using PPO to teach a single agent to balance a ball on its head.
  • Example of training Unity ML agents using the lower level Python API, using PPO implementation from OpenAI's spinningup implementation which uses MPI for parrallel training.
  • Added support for logging in tensorboard, saving experiments and configurations for experiments.
  • Also contains some legacy code for DDPG and TD3 (WIP).
  • The environment is custom built from the examples given by Unity, to include only 1 agent instead of 12 due to limitations of the python wrappers.

Ball balancing demonstration