Our Work
Robo Daycare Video Presentation
Robot Walking
Walker and Hopper was trained to walk from scratch .
The hopper was trained for 35000 episodes and walker was trained for 15000 episodes.
Robot Walking - Wind Environment
The agents were trained to walk straight and upright in a windy environment with the help of transfer learning.
Transfer learning helped in hastening the training of agents in windy environment.
Robot dribbling a ball
Transfer learning proved to be effective even for tasks like dribbling a ball.
Robot kicking a ball
With proper reward function and transfer learning the agent is able to kick the ball to a considerable distance.
Humanoid Kicking a ball
Technical Paper
Table of Contents
-
Introduction
-
Prior research
-
Tools and Software Used
-
Results and Analysis
-
Limitations , Conclusion & Future work
Link to the paper - https://github.com/csci-599-applied-ml-for-games/Robo-Daycare/blob/master/Technical_Paper.pdf