Leveraging Motion Imitation in Reinforcement Learning for Biped Character

Branches

Visualizing reference motion: demo

Train Single Motion with Mass curriculum: mass_curriculum

Train Single Motion without Mass curriculum: single_motion

Train One Policy for Multiple Motions with Skill Selector: skill_selector

Train Walking with Random Velocity and Heading: velocity_goal

Installation

please clone a branch and follow the provided instruction HOWTO to set up the workflow.

Demos

Retargetting from Deepmimic to Bob

Reference - Walking

Reference - Running

Reference - Jumping

Reference - Punching

Reference - Backflip

Training Results: One Policy per Motion

Training Results - Walk

Training Results - Walk with Random Velocity and Heading

Training Results - Run

Training Results - Jump

Training Results: Skill Selector

Training Results - Skill Selector. Only walking could be successfully imitated.