Code from "Exploring the Effectiveness of GAN-based approach and Reinforcement Learning in Character Boxing Task"
Look for papers here papers will be out soon
out downstream tasks are based on ASE and AMP.
Clone ASE repository first, then follow the instructions from its repo.
To execute Multi Agent RL environment, clone TimeChamber.
Table down below is required setting for each project. Using virtual env is recommended
| Project | Python | RL-games |
|---|---|---|
| AMP/ASE(punching, joystick) | 3.7 | 1.1.4 |
| TimeChamber(boxing) | 3.7/3.8 | 1.5.2 |
Our repository only consists of task files.
To train from scratch, see Preprocessing section.
All motion data used were from CMU We used data that were labeled as boxing and locomotion.